| # Canine OMIA ingestion | |
| VariantHound uses a retrieval-dated, checksum-pinned subset of the Online Mendelian Inheritance in Animals (OMIA) database. The transform reads the official MySQL dump as data without executing SQL and joins it to OMIA's separate internal-to-NCBI Gene export. | |
| ## Pinned inputs | |
| `data/manifests/omia-2026-09-12.json` records the byte count and SHA-256 digest of: | |
| - the complete `omia.sql.gz` dump retrieved on 2026-09-12; and | |
| - the matching `omia_gene_export.csv` mapping OMIA internal gene IDs to NCBI Gene IDs. | |
| The raw database dump is never committed or redistributed. Only the canine (`NCBI Taxonomy 9615`) normalized products are eligible for the processed Hugging Face package. OMIA requests citation of DOI `10.25910/2AMR-PV70`, an acknowledgement of the Sydney Informatics Hub, and contact about publications relying on the MySQL dump or extended OMIA data. | |
| ## Stable identity and ranking eligibility | |
| OMIA's `gene_id` is an internal identifier and is not assumed to equal NCBI Gene. `pipelines/ingest_omia.py` resolves it only through the pinned official gene export. Symbols remain display-only. Missing mappings are retained with `mapping_status: unresolved` and `rank_eligible: false`. | |
| The normalized files are: | |
| | File | Contents | | |
| |---|---| | |
| | `genes.jsonl` | Canine OMIA genes and their official NCBI Gene resolution status | | |
| | `phenes.jsonl` | Canine phenes, OMIA record IDs, inheritance, clinical text, breeds, and ontology links | | |
| | `gene_phene_evidence.jsonl` | Stable-ID gene–phene evidence with deterministic masking identifiers | | |
| | `causal_variants.jsonl` | Variants linked to canine single-locus traits, with genes, breeds, publications, HGVS fields, and stated reference assembly | | |
| | `publications.jsonl` | Publications referenced by canine phenes or causal variants | | |
| ## Leakage and temporal policy | |
| Each gene–phene and causal-variant row has a deterministic `evidence_id` and `mask_ids` containing both the evidence identifier and the binomial OMIA record identifier. Case-specific discovery evidence must be masked before feature construction, aggregation, or candidate scoring. | |
| The upstream dump is moving and has no historical release tag. This snapshot is therefore `current-snapshot-only` and cannot support temporal evaluation. A publication year is not evidence that the complete OMIA record, gene assignment, or current curation state existed in that year. | |
| ## Coordinates | |
| OMIA's variant `reference_sequence` is preserved verbatim. Recognized assembly names are copied into `assembly`; other references remain `source-reference-only`. Coordinates from different canine assemblies are never combined or lifted silently. | |
| ## Reproduction | |
| Place the two pinned inputs in `dist/omia/raw/`, then run: | |
| ```bash | |
| python3 pipelines/ingest_omia.py | |
| ``` | |
| Use `--download` to retrieve both artifacts from the official URLs. Every input checksum and every exact normalized row count must match the release descriptor or the transform stops. | |