Repository navigation
Add CARD-NLR-like system trait - #1580
Conversation
There was a problem hiding this comment.
Approving. Nothing blocking found.
What I verified. README counts match the live corpus (366 genomics, 954 total, 710 with graphs, 347 genomics PROPOSED). traitmech:000559 is the next free local ID; METPO:1051300 follows v435's METPO:1051200; METPO:1029100 is in fact the v214 placeholder for traitmech:000337, so the proposal's parent claim holds. The graph's one edge references both declared nodes and the trait node (grounding: traitmech:000559, matching identifier) is the subject, so nothing is unreachable. is a/rdfs:subClassOf is a mapped predicate in mappings/predicate_grounding.tsv. The discussion anchor causal_graphs#card_nlr_like_locus_subtype_defense names a real graph_id, which audit_discussion_anchors.py accepts. Both history/ records conform to history.yaml. Derived artifacts are regenerated, not hand-edited — and priority.json was 4 records stale on main (records: 950), so this PR also clears that backlog.
Snippet verification — honest limit. This review session had no network access, so I could not open the pinned DefenseFinder files and confirm the quotes byte-for-byte. What I could do is compare against already-merged records: the DefenseFinder_rules.tsv CARD_NLR_like row, the List_system_article.md row, the CARD_NLR__Endonuclease and CARD_NLR__Trypsin_Phospho HMM rows, and the Wein et al. sentence are all byte-identical to text already in card_nlr_system.yaml / card_nlr_gasdermin_system.yaml / card_nlr_endonuclease_system.yaml. The one row new to the corpus is CARD_NLR__Phospho_Trypsin ... | 1000 |, and its column widths (49/49/23/24/7) match the other inventory rows exactly. That is consistency, not verification — flagging so it isn't mistaken for the latter.
🔵 The label is the one thing I'd push back on. CARD-NLR-like system asserted rdfs:subClassOf CARD-NLR system reads in plain English as the "X-like is_a X" contradiction — "-like" normally excludes the thing itself. The definition and the CARD_NLR_like related synonym disambiguate it fine inside TraitMech, and the label is faithful to DefenseFinder's own subsystem name, so this is not a data defect. But the v436 proposal lifts that exact label into METPO under METPO:1029100, where the pattern is a recognised modelling smell, and the siblings use CARD-NLR <effector> system. Either rename (e.g. CARD-NLR like-subtype system) or add a line to the proposal telling the METPO editors the label is the source's subsystem string and not an "is similar to" claim.
🔵 The edge evidence snippet CARD_NLR\tCARD_NLR_like\t2\t4 is a headerless four-field prefix, but its note reads it as "two mandatory matches and four genes required." That interpretation is standard for this table and I believe it is right; a reader of the snippet alone cannot derive it. Quoting the rules-table header as a sibling evidence item would make the note self-supporting. Relatedly, the record definition captures the "two of four" mandatory requirement but drops min_genes_required = 4, so a conforming locus also needs accessory profiles — worth a clause if you touch the definition anyway.
There was a problem hiding this comment.
Approving. Identity, grounding, and evidence all hold up on the checks I could run locally.
Verified
traitmech:000559is unused elsewhere;traitmech:000337is CARD-NLR system andMETPO:1029100is the v214 placeholder for it, matching how v427/v430 cite the same parent.METPO:1051300is unused, and the proposal's claim that v435 usedMETPO:1051200is correct (Lamassu-FMO).- The graph is well-formed: both edge endpoints are declared nodes, the trait node carries the record's own identifier, one component, no orphans.
is a→rdfs:subClassOfmatchesmappings/predicate_grounding.tsv:8. - The definition tracks the rule row exactly:
Min_mandatory=2over the four-profileMandatoryset,Minimum_genes=4counting the three accessory profiles. TheCARD_NLR_likerow snippet is byte-identical to the one already merged incard_nlr_system.yaml:95. - Six of the seven remaining snippets are byte-identical to snippets already merged from the same pinned commit (Europe PMC quote: 15 occurrences;
List_system_article.mdrow: 5 records;Liste_hmm_system.mdEndonuclease/500 and Trypsin_Phospho/400; rules header). - README's regenerated table is internally consistent (REVIEWED 427, PROPOSED 439, graphs 710, total 954 all sum across rows), and the prose breakdown sums to 439. History records match the Lamassu-FMO pattern field for field.
card_nlr_system.yaml's updated discussion rationale correctly describes five CARD_NLR rule-row subtypes — that matches the five rule rows in the parent's own evidence.
One snippet I could not verify — this environment has no network access, so I could not fetch the pinned DefenseFinder files directly. Six of seven snippets were corroborated against merged records as above. The exception is the Liste_hmm_system.md row for CARD_NLR__Phospho_Trypsin (data/traits/genomics/card_nlr_like_system.yaml:37 and :135). It appears nowhere else in the corpus, and its numeric cell 1000 is the only four-digit value among the ~900 DefenseFinder HMM-inventory snippets in data/traits/. Column padding is internally consistent with the verified rows, and a high GA threshold is plausible for a fused multi-domain profile, so I am not calling this an error — but since the record itself documents that CARD_NLR__Subtilase_long_new is absent from that inventory, the inventory is demonstrably incomplete and this is the one row worth re-opening the pinned file to confirm.
🔵 The causal graph is thinner than the evidence the record already carries. card_nlr_like_locus_subtype_defense is two TRAIT nodes and a single is a edge, which restates parent_traits: and asserts nothing else. All four CARD-NLR siblings (endonuclease, gasdermin, phospho, subtilase) use a four-node GENETIC_ELEMENT locus → BIOLOGICAL_PROCESS CARD-like antiphage defense → trait shape, built from the same rules row and HMM rows this record already cites. The regenerated dashboard scores it DEEPEN_CAUSAL_GRAPH / "thin graph (1 edges) +12". docs/CURATION_PLAYBOOK.md:15 asks for a graph only where there is source-backed mechanism structure; the family shape is available here at no extra evidence cost. Lamassu-FMO (#1577) set the two-node precedent, so this is consistent with the immediately preceding PR — just not with the family it joins.
Optional and not worth a round trip: definition_source here is the rules TSV rather than DOI:10.1101/2023.05.28.542683 as the four siblings use. The TSV is arguably the more accurate source for a definition derived from a rule row, and the DOI is still carried in the proposal TSV, so this reads as a deliberate improvement rather than drift.
Summary
traitmech:000559/ CARD-NLR-like system as a PROPOSED GENOMICS child of CARD-NLR system.proposals/metpo_traitmech_v436for placeholderMETPO:1051300.Evidence
afb0e5a8b466be53586b13266f5d38d98c3ac268.metpo_traitmech_v436proposal already existed.Validation
pytest -qpassed: 1469 passed, 2 warnings.Known validation caveats
just lintcurrently stops while buildingllvmlite==0.48.0for Python 3.13 against local LLVM 20; direct Ruff on the new writer passed.ruff check src scripts testsstill hits the existing lint baseline across unrelated files.data/embeddings/was left untouched.