varianthound-data / docs /PHENIO.md
Tinyants21's picture
Publish VariantHound dataset 0.8.0 from ac278a1fef6d22e3e6f803449cc71d971ad78117
f24acd6 verified
|
Raw History Blame Contribute Delete
2.49 kB
# PHENIO / uPheno ingestion
VariantHound uses PHENIO as the cross-species phenotype backbone for deterministic semantic similarity. PHENIO is an application ontology that includes uPheno and multiple species phenotype ontologies; VariantHound does not copy the complete integrated graph into its public dataset.
## Pinned inputs
`data/manifests/phenio-2026-09-01.json` pins PHENIO release `v2026-09-01` by tag, source commit, immutable GitHub asset IDs, byte counts, and SHA-256 digests. The transform consumes:
- `phenio-relation-graph.gz`, the release's materialized relation graph; and
- `phenio-upstream-versions.tsv`, which records the integrated ontology versions.
The relation graph was generated upstream with equivalence-as-subclass, subclass output, and reflexive subclass relations. VariantHound preserves that meaning as an `entailed-reflexive-subclass-closure` rather than treating the rows as asserted direct edges.
## Compact output
`pipelines/ingest_phenio.py` scans all 82,128,289 relations but retains only the 3,972,487 subclass-closure rows needed for human/mouse cross-species comparison:
| Subject | Retained ancestors |
|---|---|
| `HP` | `HP`, `UPHENO` |
| `MP` | `MP`, `UPHENO` |
| `UPHENO` | `UPHENO` |
The output `phenotype_closure.tsv.gz` contains `phenotype_id` and `ancestor_id`. It includes one reflexive row for each retained phenotype, enabling exact self-matches. `upstream_versions.tsv` keeps the release-level provenance for all integrated sources. Labels and asserted local hierarchies remain in the separately pinned HPO and MGI snapshots.
Relations to anatomy, qualities, diseases, genes, and phenotype ontologies not used by VariantHound's first ranking path are excluded. This keeps the processed package compact without replacing PHENIO's inferred cross-species bridge with a hand-authored mapping.
## Evaluation scope
This is a current release backbone, not a historical snapshot. It may be used for current retrieval and discovery simulation when case-specific evidence is masked before feature construction, but it cannot support temporal evaluation of older cases.
## Reproduction
Place the two pinned assets in `dist/phenio/raw/`, then run:
```bash
python3 pipelines/ingest_phenio.py
```
Use `--download` to fetch the immutable release assets. Every source checksum, relation count, prefix-pair count, reflexive-term count, and output checksum is verified before the snapshot is accepted.