Add UPheno human-mouse phenotype homology ingest

[cbizon]

Summary

• Add a UPheno OBO-based ingest that infers HP-to-MP biolink:homologous_to edges from shared species-independent UPheno parents.
• Register the source and include it in RoboMouseKG.
• Keep the RoboMouse source lenient for now because strict Node Normalizer handling drops almost all MP-backed edges.

Hold

• Draft PR: hold off on merging until MP is built into Node Normalizer.
• The completed ingest produces 2,169 source edges, but strict normalization leaves only 13 normalized edges because MP identifiers mostly do not normalize today.

Verification

• uv run pytest tests/test_upheno_loader.py tests/test_graph_spec.py tests/test_source_metadata.py
• uv run pytest
• Full source ingest completed: UPhenoHumanMousePhenotypeHomology release 1315ce70e57e81f9

[eKathleenCarter]

Merged master into this branch, resolved conflicts. will do a more formal review now that the branch is up to date.

@EvanDietzMorris worth checking that I got the new
robokop-graph-spec.yaml
data_sources.py

correct.

[eKathleenCarter]

NCATSTranslator/Babel#886

was merged into main. Gaurav is currently running the Babel pipeline and will then update NN. I will keep this PR up to date on progress.

[eKathleenCarter]

I ran the parser end-to-end (orion-build UPhenoHumanMousePhenotypeHomology, then a small custom graph spec mirroring RoboMouseKG's settings) and checked it against the real upheno.obo release (108MB, 193,295 terms).

Two things I found:

1. Obsolete OBO terms were included in homology inference

parse_obo_term/parse_data never checked for is_obsolete: true. Downloading the real upheno.obo and cross-referencing obsolete HP/MP terms against their is_a lines showed 11 obsolete terms (e.g. HP:0003063, replaced by HP:0031095) still carry a live is_a: to their old UPHENO parent.

Those terms were being pulled into species_a_by_parent/species_b_by_parent and asserted as biolink:homologous_to, which introduced edges anchored to deprecated, non-canonical identifiers into the graph.

I changed parse_obo_term to record is_obsolete, and parse_data now skips obsolete terms before inference. All 11 obsolete HP/MP terms with a stale is_a are now excluded; in this particular release only one of them (HP:0007543, sharing a parent with MP:0001242) had actually produced a spurious edge, so the fixed output has exactly 1 fewer node and 1 fewer edge, but the other 10 are also no longer silently exposed to becoming spurious edges the next time UPheno adds a sibling term under their parent. Also, I added a test in tests/test_upheno_loader.py.

2. The "skip already-asserted homology" dedup was never used.

DEFAULT_EXISTING_HOMOLOGY_PREDICATES (biolink:homologous_to, RO:HOM0000007, homologous_to) was meant to skip inferring a pair the ontology already asserts homologous via an explicit relationship: line.

Grepping all 51,215 relationship: lines in the real file, and separately checking for any relationship: line connecting an HP term directly to an MP term under any predicate, found zero matches. UPheno doesn't assert cross-species homology that way.

The dedup path was dead code and only exercised by the synthetic unit-test fixture. In production, it would always report 0 "skipped" edges, implying that pre-asserted homology was being deduplicated when the mechanism was never reachable.

I removed the dead code and updated tests/test_upheno_loader.py accordingly.

[eKathleenCarter]

@cbizon and @EvanDietzMorris take a look at the changes I made, but if this all sounds good, then this parser is GTG. We should probably wait until Babel and NN finish the update to merge, though.

[eKathleenCarter]

Please see comment for requested changes.

[EvanDietzMorris]

Looks good and all MP terms normalize successfully as of Babel 2026jul22.

I made a few small changes:

• added names to the nodes so they would be there for nodes that did not normalize, but that's kind of obsolete now.
• removed this from the RESOURCE_HOGS list, which is only for sources that require lots of memory
• removed the supporting data source edge property which was always set to the same infores as the primary knowledge source