Materialize gene→taxon edges as in_taxon node properties

[kevinschaper]

Summary

Phenio expresses gene taxon as an OWL existential restriction (?gene rdfs:subClassOf [owl:onProperty RO:0002162 ; owl:someValuesFrom ?taxon]), which KGX serializes as a separate edge with relation = "RO:0002162". For gene records we want taxon as a node property (matching the canonical shape from hgnc-ingest / ncbi-gene), not a separate edge.

This adds a materialize_gene_taxon step in PhenioTransform.parse() between KGX TSV emission and the Koza enrichment. It:

• Reads PhenioTransform_edges.tsv, finds rows where relation == "RO:0002162" and subject is an identifiers.org/{hgnc,ncbigene}/ URI.
• Builds a {gene → taxon} lookup, attaches in_taxon and in_taxon_label columns to PhenioTransform_nodes.tsv (label looked up against the NCBITaxon node rows in the same file).
• Drops the materialized edges from the edges TSV.

phenio_node_sources.yaml learns about the two new columns; phenio_node_sources.py populates the corresponding biolink Gene class slots.

Context

Part of the broader pipeline cleanup tracked in monarch-initiative/monarch-app#1307. With this in place, kg-phenio's gene nodes carry the taxon information they've always semantically had but never surfaced as a queryable property — letting downstream consumers (and monarch-ingest specifically) treat phenio-derived gene nodes the same way as canonical hgnc-ingest / ncbi-gene records.

Dependencies

• For HGNC: depends on Populate in_taxon and description on HGNC gene imports monarch-initiative/phenio#130, which adds the in_taxon OWL restriction to construct-hgnc.sparql. Without that PR, phenio.json carries no taxon edges for HGNC subjects so the materializer has nothing to fold for HGNC genes (NCBIGene works either way — its CONSTRUCT already emits the restriction).
• For NCBIGene to fully work end-to-end: depends on Fix obograph_source prefix-map destruction and predicate mapping biolink/kgx#548 (TsvSource prefix-map fix). Until that's published, NCBIGene URI subjects still get filtered by BAD_PREFIXES's http/https entries before the materializer runs. After kgx#548 lands and a release is published, the URL→CURIE patches in phenio_node_sources.py:39-42 / phenio_edge_sources.py:57-58 and the http/https entries in BAD_PREFIXES become removable as a follow-up.

Test plan

• End-to-end verified on synthetic input: KGX produces PhenioTransform_*.tsv from a tiny obojson (HGNC:5 → NCBITaxon:9606, NCBIGene:698782 → NCBITaxon:9544 via RO:0002162); materializer folds in the taxon; Koza node enrichment emits HGNC:5 | biolink:Gene | A1BG | infores:hgnc | … | NCBITaxon:9606 | Homo sapiens.
• Full kg-phenio build against current phenio.json (will run as part of normal release pipeline). Expected: ncbigene gene nodes — currently filtered out by the asymmetric prefix handling — would now carry their taxon if the kgx fix were in place. With current kgx, the materializer is a no-op for any gene whose subject URI is http://... (waiting on kgx#548).

[kevinschaper]

Pushed a fixup commit (957c5d8) with three follow-ups surfaced when running PhenioTransform end-to-end against a real phenio.json + a kgx checkout with biolink/kgx#548 applied:

• sources.py: add NCBIGene to EDGE_SOURCES and NODE_SOURCES. NCBIGene URIs were always being dropped by BAD_PREFIXES on the subject side before, so the maps never had to know about them.
• phenio_node_sources.py: defensive .get() for category_sources / infores_sources. With kgx now contracting more URIs (MAXO, dct, prov, GOREL, UBERON_CORE …), unknown prefixes were KeyError-ing the whole transform.
• phenio_transform.py: the materializer's subject filter accepts both CURIE and URI form (works pre- and post-kgx#548); always emits in_taxon/in_taxon_label columns even when no rows have a value, since the koza yaml declares them.

End-to-end verified locally: 753 of 769 NCBIGene nodes get in_taxon populated; sample NCBIGene:698782 (macaque MLH1) lands as biolink:Gene | MLH1 | infores:ncbi-gene | NCBITaxon:9544 | Macaca mulatta. HGNC nodes pass through cleanly but without taxon — that needs phenio#130 to ship in a phenio rebuild first.