Address kg-model-review findings: Abscess + BacDive dual-edge + Rhea→EC

Three fixes for the modeling violations the merged-kg review surfaced.
After re-running the bacdive + rhea_mappings transforms and re-merging,
expected impact is ~133K of the 149,388 domain/range warnings cleared
(89%) without minting any new METPO term.

Finding #1 — Abscess mapping pointed at a UBERON ID that does not exist
(mappings/isolation_source_to_ontology.tsv:4)
The earlier codex_review_v2 fix swapped MONDO:0005227 for UBERON:0006548,
but UBERON has no generic "abscess" term — only as descriptive text on
unrelated anatomy entries. UBERON:0006548 is fictitious. Replaced with
HP:0025615 (HP "Abscess", a phenotypic feature defined as 'a localized
collection of purulent material surrounded by inflammation and
granulation'). HP is present in data/transformed/ontologies/hp_nodes.tsv
and avoids the disease-class concern that motivated the original fix.

Finding #2 — BacDive dual-edge assay pattern violates predicate ranges
(kg_microbe/transform_utils/bacdive/bacdive.py:2794-2812 + constants.py:295)
bacdive.py emits two edges per assay result: organism→ChEBI substrate
(the trait claim) and organism→assay (the test that produced it). Both
were emitted with the same trait predicate (METPO:2000002 assimilates,
METPO:2000011 ferments, etc.), generating 125,119 organism→Procedure
violations because those predicates' biolink-mapped range is
biolink:ChemicalEntity, not biolink:Procedure. Two changes:

* The organism→assay edge now uses METPO:2000511 (has observation), the
  existing parent of the has-X-observation predicates. This correctly
  reads as "organism has the observation that was recorded on this assay
  procedure." Substrate edge predicate is unchanged.

* ASSAY_CATEGORY swapped from "biolink:Procedure" to multi-cat
  "biolink:Procedure|METPO:1001000" (observation). biolink:Procedure
  carries the biolink semantic for downstream tooling; METPO:1001000
  satisfies METPO:2000511's existing range without any upstream METPO
  range modification. Avoids minting a new METPO term and keeps the
  proposal TSV (metpo_proposal_2026_05) untouched.

Finding #3 — Rhea→EC enabled_by violates biolink range
(kg_microbe/transform_utils/constants.py:237)
RHEA reactions (biolink:MolecularActivity) → EC enzymes (also typed
biolink:MolecularActivity in this graph) was emitted with biolink:enabled_by
whose range is biolink:PhysicalEntity, generating 7,902 violations.
Could not change EC_CATEGORY without breaking the upstream
"Protein → enables → EC" semantics that requires EC to be activity. So
swapped RHEA_TO_EC_EDGE to biolink:close_match — biolink's purpose-built
predicate for cross-vocabulary alignments, which is what Rhea↔EC actually
is (both classify the same enzymatic reaction at different granularities).
Domain/range unconstrained, semantic match is good.

Re-run scope before next merge:
* bacdive  — picks up bacdive.py edge change, ASSAY_CATEGORY multi-cat,
  and the load_isolation_source_mappings() reload of the Abscess fix.
* rhea_mappings — picks up RHEA_TO_EC_EDGE constant.
* merge — produces fresh merged-kg.tar.gz with all three fixes applied.

Validation: ROBOT template/merge/ELK reason clean; isolation_source
validator OK; pytest tests/test_extract_metpo_proposals.py +
tests/test_isolation_source_mapping_utils.py → 10 passed.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: realmarcin

kg_microbe/transform_utils/bacdive/bacdive.py

@@ -2792,18 +2792,26 @@ def run(self, data_file: Union[Optional[Path], Optional[str]] = None, show_statu
                                             )
 
                                         # NEW: Add organism → assay edge (dual-edge pattern)
-                                        # This is in addition to the direct organism→ChEBI edges above
+                                        # This is in addition to the direct organism→ChEBI edges above.
                                         # Build assay ID: assay:{kit_name}_{well_name}
                                         assay_id = f"{ASSAY_PREFIX}{assay_name}_{test_label}".replace(" ", "_")
 
-                                        # Write organism→assay edge using same METPO predicate
+                                        # The assay node carries provenance for the trait claim, not the
+                                        # trait itself, so it must NOT use the trait predicate (ferments /
+                                        # assimilates etc.) — that would map an organism to a procedure as
+                                        # if it were a substrate, violating the predicate's biolink-mapped
+                                        # range. Use METPO:2000511 (has observation); the assay nodes carry
+                                        # METPO:1001000 (observation) in their multi-category (see
+                                        # ASSAY_CATEGORY in constants.py) so the predicate's existing
+                                        # METPO:1001000 range is satisfied without an upstream change.
+                                        assay_predicate = "METPO:2000511"
                                         knowledge_level, agent_type = self._add_edge_metadata(
-                                            metpo_predicate, "RO:0002434", assay_id
+                                            assay_predicate, "RO:0002434", assay_id
                                         )
                                         edge_writer.writerow(
                                             [
                                                 organism_id,
-                                                metpo_predicate,
+                                                assay_predicate,
                                                 assay_id,
                                                 "RO:0002434",
                                                 self.knowledge_source,

kg_microbe/transform_utils/constants.py

@@ -234,7 +234,7 @@
 SUBSTRATE_TO_ASSAY_EDGE = "biolink:occurs_in"  # [substrate -> assay]
 ENZYME_TO_SUBSTRATE_EDGE = "biolink:has_input"  # [enzyme -> substrate]
 NCBI_TO_SUBSTRATE_EDGE = "biolink:consumes"
-RHEA_TO_EC_EDGE = "biolink:enabled_by"
+RHEA_TO_EC_EDGE = "biolink:close_match"
 
 # Assay → Entity predicates (methodological reference edges)
 ASSAY_HAS_OUTPUT_PREDICATE = "biolink:has_output"  # [assay -> GO/EC]
@@ -292,7 +292,12 @@
 GENOME_CATEGORY = "biolink:Genome"
 
 # Procedure categories
-ASSAY_CATEGORY = "biolink:Procedure"  # API kit assay tests
+# Multi-cat: biolink:Procedure carries the biolink semantic for downstream
+# tooling (Procedure is biolink's closest match for a microbial test kit / well),
+# and METPO:1001000 (observation) makes the node a valid object for the
+# METPO:2000511 (has observation) predicate that BacDive uses on the
+# organism→assay edge — so no METPO range-modification is needed upstream.
+ASSAY_CATEGORY = "biolink:Procedure|METPO:1001000"  # API kit assay tests
 
 # Deprecated categories - do not use
 # CHEMICAL_SUBSTANCE_CATEGORY = "biolink:ChemicalSubstance"  # removed from biolink; use CHEBI_CATEGORY

mappings/isolation_source_to_ontology.tsv

@@ -1,7 +1,7 @@
 subject_label	subject_label_normalized	object_id	object_label	object_source	predicate_id	confidence	mapping_justification	curator	source_dataset	notes	verified_date
 Abdomen	abdomen	UBERON:0000916	abdomen	UBERON	skos:exactMatch	high	semapv:LexicalMatching	ols4_auto	bacdive	The subdivision of the vertebrate body between the thorax and pelvis. The ventral part of the abdomen contains the abdom	2026-05-01
 Abort	abort							codex_review_fix_v2	bacdive	fix(codex-v2): was MONDO:0041526 (disease); abortion-as-event has no clean isolation-source ontology	2026-05-02
-Abscess	abscess	UBERON:0006548	abscess	UBERON	skos:exactMatch	high	semapv:ManualMappingCuration	codex_review_fix_v2	bacdive	fix(codex-v2): was MONDO:0005227 (disease); UBERON has abscess as tissue/structure	2026-05-02
+Abscess	abscess	HP:0025615	Abscess	HP	skos:closeMatch	high	semapv:ManualMappingCuration	kg_review_fix	bacdive	fix(kg-review): was UBERON:0006548 which does not exist in current UBERON; UBERON has no generic 'abscess' term. HP:0025615 (phenotypic feature: 'a localized collection of purulent material surrounded by inflammation and granulation') is the closest available term and verified present in data/transformed/ontologies/hp_nodes.tsv	2026-05-02
 Acidic	acidic	PATO:0001429	acidic	PATO	skos:exactMatch	high	semapv:LexicalMatching	ols4_auto	bacdive	An medium acidity quality inhering in a solution by virtue of the bearer's a high concentration of H+ ions.	2026-05-01
 Activated-sludge	activated sludge	ENVO:00002046	activated sludge	ENVO	skos:exactMatch	high	semapv:LexicalMatching	ols4_auto	bacdive		2026-05-01
 Agriculture	agriculture	ENVO:01001442	agriculture	ENVO	skos:exactMatch	high	semapv:LexicalMatching	ols4_auto	bacdive	A land use process during which terrestrial environments are modified such that they can grow crop plants or allow the r	2026-05-01