Conversation
9e65dcb to
3da09b3
Compare
…lter action - Register significance band scores and trust weight in ARAX_ranker - Add _get_significance_qualifier_value() to read from edge.qualifiers - Integrate qualifier as additive signal in edge_attribute_score_combiner - Add remove_edges_by_statistical_significance action to KG filter - Implement ordinal threshold logic in Filter_KG/remove_edges.py
…lifier - Score mapping, qualifier lookup, and edge-case tests for ranker - Ordinal filter, threshold boundary, and registration tests for filter_kg - Two @pytest.mark.slow integration tests (expand → filter/rank → resultify)
fac2c03 to
c915b7d
Compare
|
Hi @SkyeAv, this got pushed to the back-burner due to no reviewers assigned. Is this still relevant and accurate? Also, for context, do such statistical significance qualifiers show up in the translator tier0 graph, or just the multiomics/pharmacophenomics/clinical trials/etc. KP? If such edges are not in tier0 KG, this code likely would never trigger/be used |
|
@dkoslicki no CURRENT tier0 graph uses the qualifier. However, MultiomicsKG and the WellnessKG will be both ingested as a tier0 graph in the NEAR FUTURE and this PR was my attempt of preparing RTX for that. I'll add that Matt, Sierra, and I envision this qualifier becoming standard. |
|
Gotcha, thanks for clarifying. I'm seeing that the CI/CD tests haven't been run on this PR (don't know why they didn't automatically trigger). Can you confirm that all the tests pass locally? |
|
/deploy |
|
❌ ARAX preview not deployed: fork PRs are not supported (CICD-Dockerfile clones by branch name from RTXteam/RTX) |
|
@bazarkua any idea if we can get your fancy |
|
@dkoslicki I'm confirming all tests were able to pass locally. I used databases I downloaded per the wiki with the help of Steve Ramsey and Frankie Hodges. |
|
I can change |
|
/deploy |
This comment has been minimized.
This comment has been minimized.
ARAX preview✅ deployed URL: https://cicd.rtx.ai/2859/ Build: Deploy: full rebuild (--force was given) Smoke test
Comment |
Preview pytestBuild: Container: pytest: 160 passed, 0 failed, 146 skipped in 129.82s (0:02:09) |
Preview live queriesBuild: Endpoint:
|
3 participants
Adds ranking and KG-level filtering support for the
biolink:statistical_significance_qualifier(biolink/biolink-model#1766), which categorizes an edge's evidence into coarse significance bands (very_strongly_significantthroughnot_significant) as a companion to the authoritative numericp value/adjusted p valueslots. Closes #2858.Ranker
edge_attribute_score_combiner()inARAX_ranker.pynow injects a significance-band score into the existing loop algorithm (W_r = W_r + (1 - W_r) * W_i) alongside numeric p-value and other attributes — the qualifier can only boost, never penalize._significance_trust_weightconstant at0.5, with band scores ranging0.0–0.70(max effective contribution0.35). TODO comment marks this for revisit once adoption is widespread._get_significance_qualifier_value(self: Self, edge: Edge) -> Optional[str]to look up the qualifier before thefloat()conversion inedge_attribute_score_normalizer()(which would otherwiseValueErroron enum strings and silently return0).edge.qualifiers(the TRAPI path for biolink qualifier descendants), stripping anybiolink:prefix from the value.KG Filter
remove_edges_by_statistical_significanceinARAX_filter_kg.py+Filter_KG/remove_edges.py— an ordinal threshold filter, not exact-match.minimum_significance=significantremovessuggestiveandnot_significantedges while keepingsignificantand above.filter_kg(action=remove_edges_by_statistical_significance, minimum_significance=suggestive)removes onlynot_significant; supportsremove_connected_nodes,qedge_keys, andqnode_keys.allowable_actions,command_definitions, and parameter descriptors matching the existingremove_edges_by_discrete_attributepattern.Design
edge.qualifiers[]. It is read only from there — it is not registered as an edge attribute in the decorator and is not read fromedge.attributesanywhere.ARAX_filter.py) — edge-level KG filtering is the right granularity. Significance-aware pathfinding is a candidate for future work.Testing
ARAX_ranker.py— score mapping, qualifier lookup (edge.qualifiers/prefixed), edge cases (multiple qualifiers, empty lists, unrecognized values), additive-with-pValue → all PASSFilter_KG/remove_edges.py— ordinal removal, threshold boundaries (very_strongly_significantremoves 4/5,not_significantremoves 0/5), qualifier-less edges kept,command_definitionsregistration → all PASS@pytest.mark.slow) included that query liveinfores:retrieverend-to-end; these are forward-looking until BioPack-team/retriever#197 ships.