[gingin77]

Adds a new entry for premature ovarian failure 19, a recessive form of early ovarian failure caused by biallelic variants in HSF2BP. HSF2BP helps BRCA2 load the recombinases RAD51 and DMC1 onto DNA breaks during meiosis. Only three families are reported. The entry follows the existing Premature_Ovarian_Failure_20 (MSH4) entry and conforms to the meiotic_prophase_failure module.

Closes #15074

What the entry says

• Human cases.
  • A consanguineous family with three sisters homozygous for p.Ser167Leu (PMID:32845237).
  • Two unrelated women, homozygous for p.Cys128Arg and p.Leu186Pro, found in a 1,030-patient POI cohort (PMID:35174157).
  • All five had a normal menarche, irregular cycles, loss of menses between their late teens and mid-twenties, and high FSH and LH.
• Mechanism. The causal chain is mostly supported by mouse data:

HSF2BP partial loss of function (less HSF2BP to bind BRCA2)
  -> fewer RAD51/DMC1 foci and fewer crossovers in oocytes
  -> oocyte loss before primordial follicles form
  -> follicle depletion
  -> POI, secondary amenorrhea, oligomenorrhea, high FSH/LH, low estradiol, infertility

• Mouse models.
  • The Hsf2bp knockout reproduces the recombination defect and the follicle loss, although knockout females stay fertile.
  • The knock-in carrying the patients' S167L allele shows the recombination defect but keeps a normal follicle count. It is linked to the follicle-depletion node as FAILS_TO_RECAPITULATE, with a typed species-mismatch divergence.

Points for the reviewer

• The cohort paper (PMID:35174157) contradicts itself in two places. Its text calls both variants VUS, while its Table 1 lists them as likely pathogenic. It also gives patient POI-2's age when menses stopped as 18 in the text and 16 in Table 1. The entry records both readings rather than picking one.
• The first deep-research report described the wrong disease. The falcon run fell back to openscientist, whose report never mentions HSF2BP and describes ZSWIM7 disease instead. That disease is already curated as ovarian dysgenesis 10 in 46_XX_Gonadal_Dysgenesis. just preflight-dr failed it, so it is not committed.
• The committed report adds no facts beyond the papers. It is a claude_code report that passes preflight, but the agent could not fetch the primary papers. All evidence comes from the cached full texts of the two human papers and the mouse papers.
• HSF2BP's molecular function is bound as protein binding (GO:0005515). That is the only molecular function GO annotates HSF2BP with, and GO has no BRCA2-binding term, so the partner is named in preferred_term.
• What stays out. No affected man has been reported, so the entry has no male phenotype. One sister's short stature is not recorded as a disease feature, because her sisters do not share it.
• Uncited edges. The edges from follicle depletion to the hormone and menstrual phenotypes carry no citation; they are standard POI physiology.

Files

• kb/disorders/Premature_Ovarian_Failure_19.yaml is added, and stubs/Premature_Ovarian_Failure_19.yaml is deleted.
• A history/ record, the deep-research report, and the four new reference caches the entry cites. It also cites PMID:39652037, which is already cached on main.
• New term-cache rows for hgnc:5226, GO:0000730, HP:0008214 and MONDO:0030985.

Validation

All pass:

• just validate-disorders on the entry
• just validate-terms
• just count-verified-snippets (38/38)
• just validate-history
• check-entity-refs and check-causal-targets
• the gene-activity-grounding gate
• the offline snippet, title, grading, hyphen, retired-prose, block-scalar and coarse-phenotype gates

🤖 Generated with Claude Code

https://claude.ai/code/session_01H9vdyoU3AzGaJ2xhWTdPKA

[ai4c-reviewer]

Claude finished @gingin77's task in 1m 29s —— View job

---

I approved PR #15091 (premature ovarian failure 19, HSF2BP)

I found nothing that needs fixing before merge, only four small suggestions (below). The full review is in the formal review.

• Schema validation would pass
• All PMIDs verified as real: all six cited PMIDs are cached, and the titles match
• Evidence snippets are exact quotes: just count-verified-snippets verified 38 of 38
• Ontology terms correct and specific
• evidence_source classifications correct
• No pathophysiology bundling

Scope: every one of the 17 changed files belongs to this entry. The cache changes are 7 added rows, no existing rows change, and the stub is deleted.

Deep-research report: it is a real run but thin: the agent could not fetch the primary papers, so it found only one family. The YAML goes beyond it everywhere, so nothing in the report was left out.

GeneReviews: just check-genereviews finds no chapter for this disease, which confirms what the entry's note says.

Suggestions (none block merging)

1. references_cache/PMID_34794894.md (a 2022 review by França & Mendonca) is added but not cited anywhere. Cite it or drop the file.
2. The hormone replacement therapy snippet is a fragment from the guideline's "what is known already" section, and its "these effects" points back to an earlier sentence. If you keep it, quote_role: BACKGROUND fits better than REVIEW_SYNTHESIS.
3. The premature ovarian insufficiency phenotype has onset_category: YOUNG_ADULT, but patient POI-2's periods stopped at 16 or 18 (the paper gives both).
4. Optionally, add genetic counselling (NCIT:C15240).

[ai4c-reviewer]

Review: Premature Ovarian Failure 19 (HSF2BP)

The entry is ready to merge. It is careful, well sourced, and goes well beyond its deep-research report. I found no blocking issues, only the small suggestions listed below.

• Schema validation would pass (author ran validate-disorders; not second-guessed)
• All PMIDs verified as real (all 6 cited PMIDs are cached; titles match the cache)
• Evidence snippets are exact quotes (just count-verified-snippets: 38/38)
• Ontology terms correct and specific
• evidence_source classifications correct
• No pathophysiology bundling (atomic entries only)

What I checked

• Scope. All 17 files belong to this entry. The only cache changes are 7 additive rows, for the new terms only. No reversions, and the stub is deleted.
• Evidence and titles. Every new evidence item has a reference_title that matches the cached title. The mouse papers (PMID:30760716, PMID:31242413) are graded MODEL_ORGANISM or IN_VITRO as appropriate, and the human statements from PMID:32845237 are graded HUMAN_CLINICAL. The cohort paper's Table 1 rows are quoted for the hormone values, and its internal contradictions (VUS vs LP; POI-2's age when menses stopped, 18 vs 16) are recorded rather than resolved silently. That is the right call.
• Pathograph. The chain runs HSF2BP partial loss of function → impaired recombinase loading and crossover formation → oocyte elimination → follicle depletion → 7 phenotypes. Each node is atomic, conforms to the matching meiotic_prophase_failure node, and has a sensible biological_scale. Not binding an apoptosis term, because no assay showed apoptosis, is correct.
• Models. The S167L knock-in is linked to follicle depletion as FAILS_TO_RECAPITULATE, with limitations, evidence and a typed SPECIES_MISMATCH divergence. The knockout's follicle link is PARTIALLY_RECAPITULATES. This is exactly the negative-result modelling the schema exists for.
• GeneReviews. just check-genereviews reports NO_CHAPTER, and synonyms are present, so the entry's note that no GeneReviews chapter exists is verified.

Deep-research cross-check

The report (research/...-claude_code.md) is a real run with frontmatter, but it is thin: the agent could not fetch the primary papers and found only one family. The YAML goes beyond it on every axis (three families, hormone values, both mouse lines).

1. Phenotypes: adequate. Everything the report lists is present, plus estradiol and LH.
2. Subtypes: N/A.
3. Pathophysiology: adequate. The report's mention of Nat Commun 2021 (deleting the HSF2BP-binding region of BRCA2 does not reproduce the null) is optional context, not blocking.
4. Treatments: adequate for a disease with no specific therapy. HRT is present; genetic counselling, which the report suggests, is optional.
5. Genetics: adequate. All three alleles are given with HGVS and functional data. No ClinGen record exists to copy.
6. Diagnostics: adequate (exome sequencing).
7. References: the papers are well used.
8. Overall: not under-consumed.

🔵 Suggestions (non-blocking)

1. One reference cache is not cited. references_cache/PMID_34794894.md (França & Mendonca 2022 review) is added but is not cited anywhere in the entry or the report. The PR body says the five caches are the ones the entry cites, but the entry cites four of them plus PMID:39652037, which is already on main. Either cite it or drop the file.
2. The HRT snippet is a background fragment. "Although hormone therapy (HT) can mitigate some of these effects" comes from the guideline's WHAT IS KNOWN ALREADY section, and "these effects" points back to a previous sentence. It verifies and supports the claim. If you keep it, quote_role: BACKGROUND describes where it sits more precisely than REVIEW_SYNTHESIS.
3. Onset category. The POI phenotype uses onset_category: YOUNG_ADULT, but POI-2's menses stopped at 16 or 18. The onset notes already say this; an adolescent/young-adult range in the description would be more accurate.
4. Optionally, add genetic counselling (NCIT:C15240) with carrier testing of siblings, as the report suggests.

None of these block merging.

[gingin77]

Thanks for the review. All four suggestions stay as they are for this PR, because any push would dismiss the approval:

• Suggestion 1 (uncited PMID_34794894.md). This is correct, and it was my slip: the self-review dropped that review's citation but I left its cache file in. It is harmless, and I have corrected the PR description. A later sweep can remove it, or it can ride the next edit to this entry.
• Suggestion 2 (HRT snippet quote_role). I kept REVIEW_SYNTHESIS to match the sibling POF20 entry, which uses the same guideline sentence. If the two entries change, they should change together.
• Suggestion 3 (onset). The onset notes already give POI-2's 16-or-18 range. YOUNG_ADULT is the category for four of the five women.
• Suggestion 4 (genetic counselling). This would be reasonable to add, but no source specific to this disease discusses it, and POF20 does not carry it either.

---

Generated by Claude Code

[gingin77]

CI failed on the "Check for newly ungrounded genes" gate. The HSF2BP node had no molecular-function term. b6a5cab fixes it:

• The fix. The node now binds GO:0005515 protein binding with modifier: DECREASED. That is the only molecular function GO annotates HSF2BP with (IPI evidence, including PMID:31242413, which the node already cites). A note on the node says why no narrower term was used.
• Same push. It also drops references_cache/PMID_34794894.md, the uncited cache from review suggestion 1. That supersedes my earlier reply on that item.

validate, validate-terms, snippets (38/38) and the gene-grounding gate all pass locally. This push dismissed the approval, so the PR needs a fresh review.

---

Generated by Claude Code

[ai4c-reviewer]

Claude finished @gingin77's task in 3m 19s —— View job

---

I approved PR #15091 again after the latest push

• Schema validation would pass
• All PMIDs verified as real
• Evidence snippets are exact quotes (just count-verified-snippets: 38/38)
• Ontology terms correct and specific
• evidence_source classifications correct
• No pathophysiology bundling (atomic entries only)

The new commit (b6a5cab) makes two changes, and both are fine:

• HSF2BP molecular function. The HSF2BP Partial Loss of Function node now binds GO:0005515 protein binding, with preferred_term: BRCA2 binding and modifier: DECREASED.
  • validate-terms passes, and the gene-activity grounding gate no longer flags this entry.
  • The note says the modifier means less HSF2BP protein, not weaker binding. That matches PMID:32845237, which reports "a reduced protein stability of the S167L variant".
  • No duplicate keys were added.
• Uncited cache removed. PMID_34794894.md is gone, which closes my earlier suggestion 1.

When I run the grounding gate locally, it lists about 387 genes in other entries. That comes from the local run, not this PR, and this entry is not among them.

Nothing blocks merging. The full review is in the formal review.

[ai4c-reviewer]

Re-review after the latest push: approved

• Schema validation would pass
• All PMIDs verified as real
• Evidence snippets are exact quotes (just count-verified-snippets: 38/38)
• Ontology terms correct and specific
• evidence_source classifications correct
• No pathophysiology bundling (atomic entries only)

The new commit (b6a5cab) makes two changes, and both are fine.

1. Molecular function on the HSF2BP Partial Loss of Function node. It binds GO:0005515 protein binding, with preferred_term: BRCA2 binding and modifier: DECREASED.
   • just validate-terms passes, and just check-gene-activity-grounding no longer lists this entry.
   • The generic term is acceptable here: GO has no BRCA2-binding term, and the new notes say so.
   • The note says the modifier reflects less HSF2BP protein, not weaker binding. The cached full text supports that: PMID:32845237 reports "a reduced protein stability of the S167L variant (Figure 10d)".
   • No duplicate keys were added (just check-duplicate-keys reports OK).
2. references_cache/PMID_34794894.md is removed. The entry and history no longer cite it, so this closes my earlier suggestion 1. The history record's details now mention both changes.

When I run check-gene-activity-grounding locally, it lists about 387 genes in other entries. That comes from the local run, not this PR. This entry is not among them.

Last round you declined my other three suggestions, and your reasons hold. Nothing blocks merging.

[ai4c-reviewer]

Claude finished @gingin77's task in 1m 22s —— View job

---

I approved PR #15091 again after the merge of main

• Schema validation would pass (just validate: passed)
• All PMIDs verified as real
• Evidence snippets are exact quotes (just count-verified-snippets: 38/38)
• Ontology terms correct and specific
• evidence_source classifications correct
• No pathophysiology bundling (atomic entries only)

The new push is only a merge of origin/main (abd8e78), and it changes none of this PR's content:

• The PR's own files have not changed. The entry, history record, reference caches and deep-research report are byte-identical to b6a5cab, which I approved earlier.
• The merge adds only this PR's files. Compared with its main parent, it adds exactly the same 16 files.
• The cache changes only add rows. No row in cache/ was removed. check-term-cache-integrity passes, and there are no conflict markers.

Main has moved again since this merge (#15078 has landed). There is no need to refresh the branch for that, because the merge queue tests the PR against the latest main.

Nothing blocks merging.

[ai4c-reviewer]

Re-approving after the merge of origin/main (abd8e78).

• Schema validation would pass (just validate: passed)
• All PMIDs verified as real
• Evidence snippets are exact quotes (38/38)
• Ontology terms correct and specific
• evidence_source classifications correct
• No pathophysiology bundling (atomic entries only)

The merge commit changes nothing in this PR's own files. The entry, its history record, the reference caches and the research report are byte-identical to b6a5cab, which I approved earlier. Against its main parent, the merge adds exactly the 16 files of this PR. The term and enum cache changes only add rows, none are removed, and check-term-cache-integrity passes. No conflict markers.

Note that main has moved again since the merge (#15078 landed). The merge queue tests against the latest main, so no new refresh is needed for that.

Nothing blocks merging.