Human Phenotype Ontology ingestion
VariantHound uses HPO as the human phenotype hierarchy and as the source of human disease and gene phenotype annotations. The production snapshot is pinned to ontology release v2026-09-01; phenotype.hpoa declares annotation version 2026-09-02.
Pinned assets
data/manifests/hpo-2026-09-01.json records the immutable GitHub release asset ID, byte count, and SHA-256 for:
hp.obo— ontology terms, synonyms, alternate IDs, obsolescence, replacements, andis_aedges;genes_to_phenotype.txt— NCBI Gene-to-HPO associations with disease and frequency context;phenotype.hpoa— disease-to-HPO annotations, including evidence, references, frequencies, and negation.
Downloads use the GitHub asset API URL (/releases/assets/{asset_id}) with Accept: application/octet-stream. The public filename URL is retained for provenance, but is not used for reproducible retrieval: GitHub release assets are replaceable and the filename URL can therefore return bytes that do not match the release asset record.
Materialization
npm run ingest:hpo -- --download
The command verifies every raw artifact before parsing and writes four deterministic files under dist/hpo/materialized/:
| File | Contents |
|---|---|
ontology_terms.jsonl |
One ontology term per line, sorted by HPO ID |
gene_phenotypes.jsonl |
Human gene-phenotype rows with NCBI Gene IDs |
disease_annotations.jsonl |
Disease-phenotype rows with explicit negated state |
snapshot.json |
Release metadata, exact counts, output byte sizes, and SHA-256 digests |
The ingest fails on checksum drift, unexpected headers, invalid stable IDs, unresolved HPO identifiers, annotation-version drift, or any exact-count change. Empty values and the HPO - sentinel become JSON null; source identifiers, evidence codes, frequencies, references, and curation strings are otherwise preserved.
Leakage boundary
This release is a current snapshot and is marked current-snapshot-only. It may be used for the normal current-data ranking mode, but it must not be used to claim historical or temporal discovery performance. A temporal benchmark requires a separately pinned HPO and annotation snapshot whose release predates the held-out case. Direct disease and evidence masks are applied after source ingestion and before feature construction.
Scope
This module normalizes the HPO layer only. It does not infer dog-to-human mappings, merge genes by symbol, or combine Monarch, MGI, or OMIA evidence. Those sources have independent pins and ingestors so their provenance and failure modes remain separable.