---
license: other
task_categories:
- text-generation
- graph-ml
- feature-extraction
language:
- en
pretty_name: UniProt Forest-of-Thought Graphification for ToricGT
---

# UniProt Forest-of-Thought Graphification for ToricGT

This repository stores Parquet-first biological graph/FoT curation artifacts for ToricGT.
The current upload is an initial deterministic graphification sample, not the final authored reasoning corpus.

Raw upstream Parquet data is not uploaded in this initial push. Local raw sources remain organized by symlink under the ToricGT data tree and require separate license review before redistribution.

## Files

- `derived/*.parquet` graphified source-data slices
- `authored/accepted/accepted_records.parquet` when authored records are present
- `authored/accepted/validation_report.json`
- `authored/records/*.json` inspectable handwritten source records
- `manifests/uniprot_fot_build_manifest.json`
- `schema/uniprot_fot_record.schema.json`
- `LOCAL_README.md`
- `authored/README.md`
- `authored/progress_log.md`
- `GOAL.md`

## Current Local Build

- Repository: `AmelieSchreiber/uniprot_fot`
- Derived sample records: `1200`
- Raw datasets scanned: `6`
- Raw Parquet files scanned: `168`
- Readable raw Parquet files: `167`
- Raw rows indexed in manifest: `99138803`

## Format

Each row contains flat loader columns plus `graph_json`, `forest_json`, `metadata_json`, hashes, split cluster metadata, and quality flags. Graph JSON records include directed nodes/edges, GFlowNet reward metadata, continuous embedding proxies, TokenGT/TropicalGT/ToricGT fields, active support nodes, and leakage-resistant split metadata.

## Safety And License Notes

Biomedical content is for model-training curation and mechanism-focused annotation reasoning, not patient-specific medical advice. Source licenses and redistribution rights must be reviewed before uploading larger raw-derived corpora.