diff --git "a/data/gse91061_bulk/study_summary.csv" "b/data/gse91061_bulk/study_summary.csv"
new file mode 100644--- /dev/null
+++ "b/data/gse91061_bulk/study_summary.csv"
@@ -0,0 +1,52 @@
+geo_accession,gse_id,cancer_type,treatment,biopsy_location,response_criteria,library_type,data_type,study_context_biomarkers_json,study_context_clinical_observables_json,study_context_clinical_covariates_json,study_context_protocol_json,study_context_metadata_json,study_context_model_feature_policy,response_binary,response_raw,tissue,visit,gsm_title,mondo_label,ncit_label,snomed_id,rxnorm_id,drug_type,drug_canonical,uberon_id,uberon_label,audit_response,biomarker_CD274_expr,biomarker_CD8A_expr,biomarker_IFNG_expr,biomarker_CXCL9_expr,biomarker_CXCL10_expr,biomarker_B2M_expr,biomarker_LAG3_expr,biomarker_HAVCR2_expr,biomarker_PDCD1_expr,gene_mapping_rate,sample_alignment_rate,expression_type,data_source,study_context_response_schema_json,n_responders,n_non_responders
+GSM2420259,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt1_Pre_AD101148-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,1484,1696,186,15959,4827,209083,512,2284,125,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420261,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt2_Pre_AD101150-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,1340,1309,128,17735,14450,267441,459,3091,142,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420264,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,-1,UNK,melanoma,Pre,Pt23_Pre_AD313075-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,UNK,1031,1356,79,3828,4514,167145,300,2546,116,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420265,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt39_Pre_AD485899-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,24,7,0,62,37,23444,0,139,2,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420267,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt59_Pre_AD823915-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,1277,2063,207,8128,5954,356076,765,4213,266,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420268,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt82_Pre_AD823914-8,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,484,794,42,6797,1983,151993,193,1157,79,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420270,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,1,PRCR,melanoma,Pre,Pt49_Pre_AD667851-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PRCR,1122,3250,186,39053,12654,426052,570,2303,315,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420272,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt66_Pre_AD667850-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,276,272,12,583,2279,130551,96,466,32,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420273,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,-1,UNK,melanoma,Pre,Pt76_Pre_AD667852-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,UNK,1955,67,0,124,120,231725,124,150,4,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420275,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt85_Pre_AD486329-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,37,5,0,58,320,107762,18,446,4,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420277,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt89_Pre_AE070951-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,180,224,13,1495,687,155447,64,593,21,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420280,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,1,PRCR,melanoma,Pre,Pt101_Pre_AD486328-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PRCR,1421,3882,247,22651,5901,395265,575,2100,848,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420282,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt47_Pre_AD506073-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,694,311,27,2608,2145,182043,54,1582,40,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420285,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt52_Pre_AD506075-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,268,291,0,650,1160,172011,73,640,18,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420286,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt67_Pre_AD506074-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,88,553,7,324,254,112555,59,147,32,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420289,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,1,PRCR,melanoma,Pre,Pt94_Pre_AD732850-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PRCR,128,619,13,735,196,75413,50,1103,47,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420292,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt77_Pre_AD733591-7,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,2564,2909,189,22399,4282,278688,340,2058,243,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420295,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt79_Pre_AD733587-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,251,521,39,3132,1496,215879,117,825,80,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420296,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt84_Pre_AD486532-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,96,9,0,10,1,24386,15,170,1,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420299,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt98_Pre_AD733586-8,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,321,133,10,840,333,70188,85,548,45,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420302,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt62_Pre_AD608303-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,270,283,22,787,462,61287,37,406,13,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420303,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt65_Pre_AD793919-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,705,1245,77,35214,6631,242878,254,1432,137,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420307,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,1,PRCR,melanoma,Pre,Pt72_Pre_AD793922-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PRCR,157,17,0,311,478,26776,6,508,10,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420310,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt78_Pre_AD467018-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,25,7,0,62,75,45578,0,134,5,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420312,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt90_Pre_AD467873-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,949,1838,118,3994,2765,204104,421,2057,150,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420313,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt92_Pre_AE134060-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,1477,3048,211,30486,14152,333957,1539,1324,372,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420318,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt103_Pre_AE134058-2,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,251,1649,69,15146,3504,423786,379,2697,322,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420324,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,1,PRCR,melanoma,Pre,Pt3_Pre_E9024733-3,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PRCR,140,100,10,738,621,67565,49,527,22,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420326,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt4_Pre_E9021023-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,187,1012,61,1290,832,141311,345,834,148,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420329,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt5_Pre_E9021022-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,501,1857,109,4326,2275,220211,546,1835,366,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420331,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt9_Pre_E9021024-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,169,356,0,210,305,54098,88,1245,2,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420333,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt10_Pre_E9047565-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,290,102,5,1321,1383,74919,72,1025,1,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420336,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt17_Pre_E9047563-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,608,12,0,77,121,51614,12,228,2,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420337,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,1,PRCR,melanoma,Pre,Pt18_Pre_E9024732-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PRCR,1203,2166,94,8065,2741,237762,384,1465,186,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420341,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt38_Pre_E9200719-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,111,66,2,209,239,67812,35,559,7,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420343,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,1,PRCR,melanoma,Pre,Pt48_Pre_E9047561-7,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PRCR,224,585,37,2140,1030,149612,167,987,36,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420344,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt36_Pre_AD467095-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,191,53,0,263,66,156205,40,4754,10,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420346,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt37_Pre_AD502452-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,152,252,12,1839,1426,157470,54,382,29,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420350,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt46_Pre_AD467096-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,1122,1022,44,4378,1241,516766,143,9229,67,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420352,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt24_Pre_AD436687-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,1021,956,63,1074,793,389080,76,10761,29,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420354,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt27_Pre_AD453873-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,170,356,51,1049,259,90476,76,1012,11,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420355,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt29_Pre_AD497504-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,401,369,18,2222,1271,116078,81,1353,28,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420357,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,1,PRCR,melanoma,Pre,Pt30_Pre_AD497503-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PRCR,117,326,14,2611,1122,38897,90,493,19,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420359,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt31_Pre_AD453872-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,582,809,32,5693,2724,266885,249,744,79,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420360,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt8_Pre_AD153354-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,255,918,18,6706,5093,224165,245,1266,70,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420362,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt11_Pre_AD153352-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,888,2326,254,18324,12797,485245,402,2461,117,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420366,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,SD,melanoma,Pre,Pt26_Pre_AD467789-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,SD,182,284,22,2491,1381,134558,134,292,6,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420368,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt28_Pre_AD297619-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,291,590,66,9130,1717,73942,87,556,22,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420370,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,1,PRCR,melanoma,Pre,Pt44_Pre_AD467790-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PRCR,705,2395,201,26392,12361,219164,871,2852,150,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420372,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,0,PD,melanoma,Pre,Pt106_Pre_AD502250-5,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PD,203,1219,55,15887,3530,212501,409,1263,157,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39
+GSM2420375,GSE91061,advanced melanoma (unresectable or metastatic),Nivolumab (anti-PD-1) 3 mg/kg IV every 2 weeks; CA209-038 clinical study (NCT01621490); cohort includes ipilimumab-naive (n=33) and ipilimumab-progressed (n=35) patients,"tumor biopsy (metastatic melanoma lesion, same site biopsied pre- and on-treatment)","RECIST v1.1 best overall response to nivolumab: CR or PR combined into PRCR = responder; PD = non-responder; SD = stable disease (intermediate phenotype per paper, typically grouped with non-responders for binary classification); UNK = unknown, exclude from training. OS data available in paper supplementary tables (not in GEO characteristics_ch1).",polyA,bulk,"[{""approval_context"": ""PD-L1 protein IHC is a biomarker for nivolumab in melanoma but is not an FDA companion diagnostic (unlike NSCLC); mRNA is a proxy for IHC protein."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""29126"", ""gene_symbol"": ""CD274"", ""name"": ""CD274 (PD-L1 mRNA)"", ""notes"": ""Paper notes CD274 mRNA was NOT significantly differentially expressed pre-therapy between responders and non-responders, but IHC PD-L1 protein correlated with response in Ipi-P patients. Include for benchmarking; expect lower AUROC than in NSCLC."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""925"", ""gene_symbol"": ""CD8A"", ""name"": ""CD8A (cytotoxic T cell marker)"", ""notes"": ""Marker of CD8+ cytotoxic T cell infiltration. Paper confirmed increase in CD8+ T cells on-therapy associated with response. Pre-therapy levels did not significantly differ between responders and non-responders in this cohort, but published data across anti-PD-1 trials support baseline CD8A as predictive."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3458"", ""gene_symbol"": ""IFNG"", ""name"": ""IFNG (IFN-gamma)"", ""notes"": ""IFN-gamma signaling pathway enriched in pre-therapy responder samples per GSEA (paper Figure S3D). Core component of the T-cell inflamed gene expression signature. Strong published evidence across multiple anti-PD-1 cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""4283"", ""gene_symbol"": ""CXCL9"", ""name"": ""CXCL9 (IFN-gamma-induced chemokine)"", ""notes"": ""IFN-gamma-inducible chemokine driving CD8+ T cell recruitment into the tumor. Well-validated predictor of anti-PD-1 response across melanoma and other tumor types. Likely differentially expressed given IFN-gamma pathway enrichment in responders."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3627"", ""gene_symbol"": ""CXCL10"", ""name"": ""CXCL10 (IP-10)"", ""notes"": ""IFN-gamma-inducible T-cell attractant; frequently co-expressed with CXCL9 as part of the inflamed TME signature. Associated with anti-PD-1 response in multiple cohorts."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""567"", ""gene_symbol"": ""B2M"", ""name"": ""B2M (beta-2-microglobulin)"", ""notes"": ""Required for HLA class I surface stability and antigen presentation. Paper reports one responder harbored B2M frameshift + LOH, associated with acquired resistance. B2M loss is a known immune escape mechanism; mRNA expression reflects antigen presentation capacity and may associate with baseline T-cell recognition potential."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""3902"", ""gene_symbol"": ""LAG3"", ""name"": ""LAG3"", ""notes"": ""Immune checkpoint receptor on exhausted T cells; paper explicitly identifies LAG3 as upregulated on-therapy (Figure 4A). Baseline LAG3 expression in pre-therapy samples reflects degree of pre-existing T cell exhaustion and co-inhibitory signaling in the TME."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""84868"", ""gene_symbol"": ""HAVCR2"", ""name"": ""HAVCR2 (TIM-3)"", ""notes"": ""TIM-3 immune checkpoint; paper identifies HAVCR2 as upregulated on-therapy in responders (Figure 4B). Baseline expression marks T-cell exhaustion and co-expression with other checkpoints. Relevant as pre-therapy exhaustion landscape marker."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Cytolytic Score (GZMA + PRF1 composite)"", ""notes"": ""Geometric mean of GZMA (Entrez 3001) and PRF1 (Entrez 5551) expression (Rooney et al. 2015). Directly cited in paper as associated with benefit to nivolumab in BOTH Ipi-N and Ipi-P cohorts (p=0.005 and p=0.043 respectively). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz \u2014 highest-confidence biomarker in this dataset."", ""source"": ""gene_proxy"", ""type"": ""composite_signature""}, {""approval_context"": ""FDA approved pembrolizumab for TMB-high solid tumors (\u226510 mut/Mb, 2020). In this study TMB associated with OS in Ipi-N but not Ipi-P patients."", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": null, ""gene_symbol"": null, ""name"": ""Tumor Mutational Burden (TMB)"", ""notes"": ""WES-derived TMB available in paper Tables S2/S3 but NOT in GEO characteristics_ch1. Cannot be extracted from RNA-seq. Would require manual merge from supplementary data. Informational only for benchmarking against RNA-seq-derived features."", ""source"": ""literature_proposed"", ""type"": ""genomic_tmb""}, {""approval_context"": """", ""characteristics_column"": null, ""clinical_standard"": false, ""entrez_id"": ""5133"", ""gene_symbol"": ""PDCD1"", ""name"": ""PDCD1 (PD-1)"", ""notes"": ""PD-1 receptor on T cells; direct pharmacological target of nivolumab. Paper reports PDCD1 upregulated on-therapy regardless of response; baseline expression was not significantly differential. Include for completeness and benchmarking; lower expected AUROC at pre-treatment timepoint."", ""source"": ""gene_proxy"", ""type"": ""gene_expression""}]","[{""characteristics_column"": ""response"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""response"", ""notes"": ""RECIST v1.1 best overall response label: PRCR (CR+PR), SD, PD, UNK. This is the primary outcome target \u2014 must not be used as a model input feature."", ""patient_level"": true, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""tissue"", ""leakage_risk"": ""low"", ""model_feature_candidate"": false, ""name"": ""tissue"", ""notes"": ""All samples are labeled 'melanoma'; zero variance across dataset. Not useful as a predictive feature."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": ""visit (pre or on treatment)"", ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""visit (pre or on treatment)"", ""notes"": ""Timepoint label ('Pre' or 'On'). Must be used to filter to pre-treatment samples only. On-treatment samples (cycle 1 day 29) contain post-treatment pharmacodynamic signal and must be excluded from predictive models."", ""patient_level"": false, ""source"": ""sample_characteristics"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""prior ipilimumab status"", ""notes"": ""Ipilimumab-naive (Ipi-N, n=33) vs ipilimumab-progressed (Ipi-P, n=35). Key stratification variable from paper Table S2. Affects baseline immune landscape, TMB predictiveness, and T-cell repertoire dynamics. Should be modeled as a covariate or used for stratified analysis."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""binary""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""M-stage"", ""notes"": ""AJCC M-stage at study entry: M0 (n=1), M1a (n=13), M1b (n=10), M1c (n=36), unknown (n=8). Available in paper Table S1. M1c predominant in cohort."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""age"", ""notes"": ""Median 55 years (range 22\u201389) from paper methods section. Available in paper Table S1 per patient."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""tumor mutation burden"", ""notes"": ""WES-derived total non-synonymous mutation count; median 183 (range 1\u20137360). Available in paper Tables S2/S3. Associated with OS and response in Ipi-N patients only."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""clonal mutation load"", ""notes"": ""Number of clonal mutations (CCF \u22650.95 CI lower bound) from WES/PyClone analysis. Improves OS prediction over total TMB in Ipi-N patients. Available in paper supplementary tables."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""melanoma mutational subtype"", ""notes"": ""TCGA-defined subtypes: BRAF-mutant, NRAS-mutant, NF1-mutant, triple wild-type (WT). Triple-WT enriched in Ipi-P cohort (57% vs 33% in Ipi-N). Available in paper Table S2."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}, {""characteristics_column"": null, ""leakage_risk"": ""low"", ""model_feature_candidate"": true, ""name"": ""cytolytic score"", ""notes"": ""Geometric mean of GZMA and PRF1 expression from RNA-seq (Rooney et al. 2015). Pre-computed score available in GEO supplementary file GSE91061_BMS038109Sample_Cytolytic_Score_20161026.txt.gz. Associated with response benefit in both Ipi-N and Ipi-P cohorts per paper."", ""patient_level"": false, ""source"": ""protocol_or_paper"", ""value_type"": ""numeric""}, {""characteristics_column"": null, ""leakage_risk"": ""high"", ""model_feature_candidate"": false, ""name"": ""on-treatment genomic contraction/persistence"", ""notes"": ""Post-treatment WES-derived molecular phenotype (genomic contraction vs persistence). Requires on-treatment biopsy \u2014 cannot be used for pre-treatment prediction. High leakage risk."", ""patient_level"": true, ""source"": ""protocol_or_paper"", ""value_type"": ""categorical""}]","[""response"", ""tissue"", ""visit (pre or on treatment)"", ""prior ipilimumab status (Ipi-N vs Ipi-P)"", ""M-stage"", ""age"", ""tumor mutation burden"", ""clonal mutation load"", ""melanoma mutational subtype (TCGA)""]","{""biospecimen_timing"": ""Pre-treatment biopsy: 1\u20137 days before first nivolumab dose. On-treatment biopsy: cycle 1, day 29 (days 23\u201329), collected from the same anatomical lesion site. Tissue stored in RNAlater (Ambion) for RNA/DNA extraction."", ""eligibility_summary"": ""Advanced (unresectable or metastatic) melanoma; ipilimumab-naive or ipilimumab-progressed; sufficient tumor tissue for paired pre- and on-treatment biopsies required; multi-institutional."", ""nct_id"": ""NCT01621490"", ""primary_outcomes"": [""Median change from baseline to week 7 in IFN and IFN-gamma inducible factors"", ""Tumor infiltrating lymphocytes (TILs) by CD8/CD4 IHC at baseline and on-treatment biopsy""], ""protocol_source"": ""ClinicalTrials.gov + paper (Riaz et al., Cell 2017, PMID 29033130)"", ""response_assessment_time"": ""RECIST v1.1 best overall response; radiographic imaging approximately every 8 weeks until progression; progression confirmed with repeat CT ~4 weeks later."", ""secondary_outcomes"": [""Transcriptomic and genomic changes pre- vs on-therapy"", ""Tumor mutation load association with OS and RECIST response"", ""TCR repertoire diversity dynamics on-therapy"", ""Clonal evolution and neoantigen depletion on-therapy"", ""Survival benefit stratified by prior ipilimumab exposure""], ""study_design"": ""observational"", ""treatment_arms"": [""Nivolumab 3 mg/kg IV q2w monotherapy (CA209-038; includes both Ipi-naive and Ipi-progressed patients)"", ""Nivolumab + Ipilimumab combination arm (separate arm within NCT01621490; not represented in GSE91061 RNA-seq data)""], ""trial_phase"": ""Phase 1""}","{""model_training_notes"": ""CRITICAL: Filter 'visit (pre or on treatment)' == 'Pre' to isolate 51 pre-treatment samples. Exclude 'UNK' response samples from training. For binary classification: PRCR=1 (responder), PD=0 (non-responder); SD is intermediate \u2014 either exclude or group with PD. Dataset contains two immunologically distinct sub-populations (Ipi-N vs Ipi-P) with different baseline immune states and different TMB-response relationships \u2014 strongly consider stratification or inclusion of Ipi status as a covariate. OS/PFS from paper supplementary tables not encoded in characteristics_ch1 and would require manual merging. Cytolytic score supplementary file can be merged with expression matrix by sample ID. GPL9052 = Illumina Genome Analyzer IIx; hg19 alignment; FPKM and raw counts both available as supplementary files. No known duplicate cohorts in public databases."", ""platforms"": [""GPL9052""], ""pmids"": [""29033130""], ""primary_accession"": ""GSE91061"", ""processing_adapter_context"": {}, ""repository"": ""geo"", ""sample_characteristics_keys"": [""response"", ""tissue"", ""visit (pre or on treatment)""], ""title"": ""Tumor and Microenvironment Evolution during Immunotherapy with Nivolumab""}",Use patient-level clinical/biomarker columns as candidate features; study_context_* columns are study-level annotations unless explicitly promoted.,1,PRCR,melanoma,Pre,Pt34_Pre_AD466985-6,metastatic melanoma,Locally Advanced Unresectable Melanoma,SCT:443493003,,monoclonal_antibody,Nivolumab,UBERON:1000010,mole,PRCR,1177,4211,190,18374,4690,586324,834,6872,197,0.9548,1.0,counts,geo_processed,"{""non_responder_values"": [""PD"", ""SD""], ""raw_column"": ""response"", ""raw_values"": [""PD"", ""PRCR"", ""SD"", ""UNK""], ""responder_values"": [""PRCR""], ""response_continuous_column"": null, ""response_notes"": ""PRCR encodes combined CR+PR as a single label. SD patients represent a molecular intermediate phenotype between PRCR and PD per paper (paper shows SD subclone dynamics differ from both responders and progressors); analysts may choose to exclude SD or group with PD. UNK must be excluded. OS and PFS data are available in paper supplementary tables (Table S2) but are NOT encoded in GEO characteristics_ch1 fields."", ""standard_tier"": ""RECIST_1.1"", ""survival_event_column"": null, ""survival_event_values"": [], ""survival_os_column"": null, ""survival_pfs_column"": null}",10,39