dataset card: Dataset Viewer configs (8 configs)

README.md

@@ -9,50 +9,94 @@ tags:
 - zero-shot
 - protein-embeddings
 size_categories:
-- n<1K
+- 1M<n<10M
+configs:
+- config_name: cohort
+  data_files:
+  - split: train
+    path: viewer/cohort/*.parquet
+- config_name: annotations
+  data_files:
+  - split: train
+    path: viewer/annotations/*.parquet
+- config_name: responses
+  data_files:
+  - split: train
+    path: viewer/responses/*.parquet
+- config_name: embeddings_esm2_650m
+  data_files:
+  - split: train
+    path: viewer/embeddings/embedding_esm2_650m.parquet
+- config_name: embeddings_esm2_3b
+  data_files:
+  - split: train
+    path: viewer/embeddings/embedding_esm2_3b.parquet
+- config_name: embeddings_protrek
+  data_files:
+  - split: train
+    path: viewer/embeddings/embedding_protrek.parquet
+- config_name: splits
+  data_files:
+  - split: train
+    path: viewer/splits/splits.parquet
+- config_name: rpe1_splits
+  data_files:
+  - split: train
+    path: viewer/splits/rpe1_splits.parquet
 ---
 
 # AMBIMOD benchmark data (K562)
 
 Derived evaluation data for the AMBIMOD zero-shot perturbation-prediction
-benchmark. ~27 MB unpacked; raw datasets are NOT included (see Raw Data Sources).
+benchmark. The data is organized into Dataset-Viewer configs (browse each
+tab above) plus a ready-to-use archive for the benchmark code.
 
 ## Dataset Access
-This repository hosts the processed, benchmark-ready derived data (490-TF
-cohort, locked response matrices, protein embeddings, split manifests).
 
-## Raw Data Sources
-- K562 GWPS + RPE1 essentialome (Perturb-seq): https://gwps.wi.mit.edu/
-- K562/RPE1 figshare mirror (RPE1 h5ad, direct download): https://figshare.com/ndownloader/files/42362275
-- Norman 2019 CRISPRa (GEARS-official file, Harvard Dataverse): https://dataverse.harvard.edu/api/access/datafile/6154020
+Two equivalent access paths:
+
+**A. Load individual tables with `datasets`** (what the Dataset Viewer shows):
 
-## Contents
+```python
+from datasets import load_dataset
 
+cohort = load_dataset("wozhen/ambimod", "cohort", split="train")
+splits = load_dataset("wozhen/ambimod", "splits", split="train")
+responses = load_dataset("wozhen/ambimod", "responses", split="train")
 ```
-cache/
-  embeddings/                 # protein embeddings for the 490-TF cohort
-    k562_esm2_650m_491x1280.npy    # ESM2-650M residue-mean (default)
-    esm2_3b_490.npy                # ESM2-3B
-    protrek_490.npy                # ProTrek-650M
-  k562_experiments/
-    cohort/cohort_doublet_filtered.csv       # 490 TFs: gene, length, sequence, UniProt
-    responses_doublet_filtered/
-      responses_log1p_delta.npy             # (490, 5, 1000) log1p-delta responses
-      target_gene_indices.npy               # 1000 target-gene indices
-      metadata.json
-    annotations/
-      dbd_annotations_doublet_filtered.csv  # DNA-binding-domain families
-      ensembl_paralogs_doublet_filtered.json# paralog graph (leakage control)
-data/
-  splits/
-    k562/{design}_seed{NN}.json  # 15 frozen split manifests (train/test TF lists)
-    eligible_tfs.json, seed*.json  # RPE1 frozen splits (raw-data pipeline)
-  embedding_checksums.txt          # SHA-256 of every embedding file
-  download_*.sh                    # raw-data download scripts (optional rebuild)
-EMBEDDINGS.md                      # embedding provenance + reproduction guide
+
+**B. Download the benchmark archive** (exact layout the code expects —
+`cache/`, `data/splits/`, checksums):
+
+```bash
+huggingface-cli download wozhen/ambimod --repo-type dataset --local-dir .
+tar xzf ambimod-benchmark-data.tar.gz
 ```
 
-### Embedding provenance
+## Configs (Dataset Viewer tabs)
+
+| Config | Rows | Columns | Content |
+|---|---|---|---|
+| `cohort` | 490 | tf, n_cells, sequence_length, sequence, uniprot_header | quality-controlled K562 TF cohort (UniProt-reviewed human sequences) |
+| `annotations` | 490+490 | tf ↔ DBD family (InterPro); tf ↔ paralog list | leakage-control annotations |
+| `responses` | 2,450,000 | tf, replicate (1–5), gene_idx, gene_id, log1p_delta | held-out evaluation responses: log1p-delta per TF × replicate × 1,000 target genes |
+| `embeddings_esm2_650m` | 627,200 | tf, dim, value | ESM2-650M embeddings (490 × 1280, L2-normalized) |
+| `embeddings_esm2_3b` | 1,254,400 | tf, dim, value | ESM2-3B embeddings (490 × 2560) |
+| `embeddings_protrek` | 501,760 | tf, dim, value | ProTrek-650M embeddings (490 × 1024) |
+| `splits` | 7,350 | design, seed, role, tf | the 15 frozen K562 splits (random / dbd_family / embed_ball × 5 seeds), train/test membership |
+| `rpe1_splits` | 6,534 | seed, role, perturbation | frozen RPE1 split manifests (reference for the raw-data pipeline) |
+
+Embeddings are stored in long format (`tf, dim, value`) for Viewer
+compatibility; the benchmark archive (`ambimod-benchmark-data.tar.gz`)
+contains the same values as dense `(490, d)` float32 arrays.
+
+## Raw Data Sources
+
+- K562 GWPS + RPE1 essentialome (Perturb-seq): https://gwps.wi.mit.edu/
+- K562/RPE1 figshare mirror (RPE1 h5ad, direct download): https://figshare.com/ndownloader/files/42362275
+- Norman 2019 CRISPRa (GEARS-official file, Harvard Dataverse): https://dataverse.harvard.edu/api/access/datafile/6154020
+
+## Embedding provenance
 
 All embedding caches were **computed by us** from public model weights applied
 to the cohort's UniProt-reviewed sequences — no third-party embedding files
@@ -60,20 +104,18 @@ are redistributed. Checkpoints used (all public):
 
 | Embedding | Checkpoint | Pooling |
 |---|---|---|
-| `k562_esm2_650m_491x1280.npy` | `facebook/esm2_t33_650M_UR50D` | residue mean (1022-aa length-weighted chunks), L2-normalized |
-| `esm2_3b_490.npy` | `facebook/esm2_t36_3B_UR50D` | same |
-| `protrek_490.npy` | `westlake-repl/ProTrek_650M` | official `get_protein_repr` |
+| `esm2_650m` | `facebook/esm2_t33_650M_UR50D` | residue mean (1022-aa length-weighted chunks), L2-normalized |
+| `esm2_3b` | `facebook/esm2_t36_3B_UR50D` | same |
+| `protrek` | `westlake-repl/ProTrek_650M` | official `get_protein_repr` |
 
 `scripts/recompute_embeddings.py` in the code repository rebuilds any cache
 from these checkpoints and fails on checksum mismatch. The RPE1 patched
 ProTrek cache and its gated sanity receipt are documented in
 `EMBEDDINGS.md` in the code repository.
 
-## Usage
+## Usage with the benchmark code
 
 ```bash
-huggingface-cli download wozhen/ambimod --repo-type dataset --local-dir .
-tar xzf ambimod-benchmark-data.tar.gz
 pip install -e git+https://github.com/ambimod/ambimod#egg=ambimod
 python -c "from mechid.benchmark import load_benchmark; s = load_benchmark(); print(len(s))"
 # -> 15