Review 2q37 microdeletion syndrome evidence

[cmungall]

Summary

• keep MONDO:0010886 scoped to the variable terminal/interstitial multigene 2q37 deletion
• remove the HDAC4-only subtype and represent HDAC4 as an incompletely penetrant skeletal contributor with explicit counterevidence
• separate deletion-negative HDAC4 loss-of-function disease and 14-3-3-site missense disease as differentials
• rebuild the causal graph around breakpoint-dependent gene content, cautious HDAC4 dosage evidence, patient-derived nuclear-architecture findings, and uncertain 2q37.1 Wilms susceptibility
• retain quantitative phenotype bands only where the 103-person review supplies a matching denominator and composite-compatible HPO term
• update diagnosis to CMA-first with selective karyotype/parental chromosome studies
• replace fixed Wilms/renal surveillance with individualized discussion plus an open knowledge gap
• remove the unsupported NCT01238250 entry and reduce the bibliography to used references
• add generated PMID caches and the required phenotype-term cache entry
• append a review history record

Disease-boundary and evidence notes

The record no longer treats intragenic HDAC4 disease as a subtype of the constitutional deletion. Positive HDAC4 evidence is balanced by normal intelligence in some individuals with HDAC4 haploinsufficiency and by intellectual/skeletal findings in a small deletion series where one deletion did not include HDAC4.

The previous syndrome-wide Wilms screening recommendation is replaced by cautious language limited to deletions encompassing 2q37.1. No interval or stopping age is encoded because the generated sources do not establish one.

The repository's generated DiseaseTerm enum does not yet include the current MONDO identities for the two HDAC4 differentials (MONDO:1060110 and MONDO:0859232), so those verified IDs are preserved in notes rather than invalid structured bindings.

Validation

• just validate-disorders kb/disorders/2q37_Microdeletion_Syndrome.yaml
• uv run python -m dismech.graph --validate kb/disorders/2q37_Microdeletion_Syndrome.yaml
• just validate-history history/disorders/2q37_Microdeletion_Syndrome/2026-07-14T225525Z-codex-review.yaml
• 51 evidence snippets independently rechecked against generated caches after whitespace normalization
• two independent agent reviews returned GO after requested semantic fixes
• pre-commit hooks pass

The repository-wide graph check still reports pre-existing integrity issues in 81 unrelated disorders; 2q37 is not among them. The whole-cache frontmatter check still reports the pre-existing PMID_31909928.md authors/journal issue; all newly added caches contain the expected generated metadata.

Related expert review: #6560
Follow-up to prior review: #3479

[ai4c-agent]

Claude finished @cmungall's task in 3m 49s —— View job

---

PR Review: 2q37 Microdeletion Syndrome ✅

Summary checklist:

• Schema validation would pass (CI-authoritative; no structural concerns)
• All PMIDs verified as real (11 references, all resolve)
• Evidence snippets are exact quotes (spot-checked against caches; line-wrapping normalized)
• Ontology terms correct and specific (17 HP terms + MONDO differentials verified)
• evidence_source classifications correct (MODEL_ORGANISM/IN_VITRO correctly split)
• No pathophysiology bundling (all 6 nodes atomic)

This is a careful, high-quality evidence re-review of an existing entry. The scoping decisions are well-reasoned and explicitly documented.

Verification performed:

• Snippets — Spot-checked all newly added caches (PMID_30848064, 24715439, 20691407, 31990478, 19789318, 29921581, 33537682, 20466091). The critical 103-person frequency snippet ("overweight and obesity (34%)…type E brachydactyly (48%)") matches PMID_30848064 lines 72–74 (folded scalar normalizes the wrap). All verified.
• Frequency bands — Only the four phenotypes with denominators in the 103-person review carry bands: facial 86% → VERY_FREQUENT, mental function 79% → FREQUENT, brachydactyly 48% → FREQUENT, weight 34% → FREQUENT. All correct; the 20%–35% seizure range is deliberately left unbanded. Good discipline.
• evidence_source — PMID:20691407 is correctly split (HUMAN_CLINICAL for the sequencing quote, MODEL_ORGANISM for the Hdac4⁻/⁻ mouse quote); PMID:29921581 patient-derived MSCs → IN_VITRO. Correct.
• Disease boundary — Removing the HDAC4-only subtype and re-representing intragenic HDAC4 LoF and the 14-3-3-site missense disorder as differentials (not subtypes) is the right lumping/splitting call. Positive HDAC4 evidence is properly balanced with counterevidence (PMID:24715439 normal intelligence; PMID:31990478 skeletal findings in an HDAC4-non-deleted case).
• Pathophysiology — All nodes atomic; hub node "Variable neurodevelopmental and congenital effects" fans out to atomic phenotype targets via downstream edges. Pathograph forms a connected component.
• GeneReviews (item 15) — PMID:20301337 tagged and actively mined across all four sections: Clinical Characteristics (dev delay, seizures, CHD, CNS, renal, GI phenotypes), Diagnosis (karyotype/rearrangement snippet), Management (multidisciplinary care, early intervention, counseling, Wilms), Genetic Counseling (de novo/5% translocation-carrier + sib-risk snippets). The retired-chapter caveat is handled correctly (PARTIAL, historical framing).
• Deep-research cross-check — Both asta and falcon artifacts present and genuine (frontmatter + 29 citations in falcon). YAML consumption is intentionally narrower and the narrowing is well-signaled in the PR body and history record. No blocking omissions.

🔵 SUGGESTION (non-blocking): The falcon DR surfaced a contemporary 2024 AACR Pediatric Cancer Working Group surveillance update (Kalish et al., PMC11611621) that lists 2q37 deletion with undefined Wilms risk and a "shared decision" recommendation — which aligns exactly with this PR's shift to individualized discussion. Citing it would strengthen both the Individualized discussion of Wilms tumor surveillance treatment and the gap_2q37_1_wilms_risk_and_surveillance knowledge gap (whose rationale currently states no current consensus recommendation is available in the caches). The history record already links PMC11611621 and defers this to expert-review issue #6560, so this is a defensible scoping choice — noting it only as an enhancement opportunity.

Verdict: No CRITICAL or IMPORTANT issues. Snippets are exact quotes, PMIDs are real, ontology terms are accurate, evidence sources are correctly classified, GeneReviews is mined across all present sections, and pathophysiology is atomic. Approving.

[ai4c-agent]

Careful, high-quality evidence re-review of an existing entry. Verified all newly added PMID caches (snippets are exact quotes; line-wrapping normalized), confirmed the four retained frequency bands map correctly to the 103-person denominators, and confirmed evidence_source splits (HUMAN_CLINICAL vs MODEL_ORGANISM mouse vs IN_VITRO patient MSCs) are correct. Disease-boundary rework (HDAC4-only subtype removed; intragenic HDAC4 LoF and 14-3-3-site missense modeled as differentials) is the right lumping/splitting call, with positive HDAC4 evidence properly balanced by counterevidence. GeneReviews (PMID:20301337) is tagged and actively mined across Clinical Characteristics, Diagnosis, Management, and Genetic Counseling. Pathophysiology nodes are atomic. Deep-research (asta + falcon) cross-check found no blocking omissions; the YAML's narrower scope is well-signaled. Only a non-blocking suggestion: the falcon DR surfaced the 2024 AACR Wilms surveillance update (Kalish, PMC11611621, 'shared decision') that aligns with this PR's individualized-discussion shift and could strengthen the Wilms treatment/knowledge-gap — already deferred to expert-review #6560. No CRITICAL or IMPORTANT issues.