Skip to content

Add NF2-Related Schwannomatosis (MONDO:0007039) - #17478

Open
cbhindex wants to merge 2 commits into
mainfrom
nbp-nf2-schwannomatosis
Open

cbhindex wants to merge 2 commits into
mainfrom
nbp-nf2-schwannomatosis

Conversation

@cbhindex

Copy link
Copy Markdown
Contributor

New entry for NF2-related schwannomatosis, the syndrome rather than its tumours.
Acoustic_Neuroma, Meningioma, Schwannoma and Ependymoma already carry the
biology of the constituent tumours, so this curates what they do not: the germline
lesion, the two-hit mechanism, the 2022 diagnostic criteria, the tumour spectrum as
phenotypes, mosaicism, surveillance and outcome.

Closes #17397.

There is no stub to delete and no research report to commit.

What it contains

  • A six-node chain: germline NF2 inactivation, a somatic second hit, loss of
    merlin-dependent growth restraint, YAP and TAZ activation, schwannoma formation,
    mass effect on the eighth nerve. Seven of ten phenotypes hang off it.
  • The second-hit node binds all three lineages the syndrome's tumours arise in —
    Schwann, meningothelial and ependymal — and both NF2-bearing nodes bind merlin's
    signalling adaptor activity.
  • Two inheritance blocks. Mosaicism is one of them rather than a footnote on the
    dominant block: a large minority of patients are mosaic, their disease is milder,
    later and more asymmetrical, and their transmission risk is below the dominant one.
  • The ClinGen Definitive gene-disease validity assertion for NF2, recorded as a
    gene_disease_validity block against this entry's own MONDO term.
  • Three discussions. One is a CONTROVERSY: two full-text papers disagree about
    where merlin acts on LATS, one placing it at the plasma membrane and the other
    concluding it acts largely if not exclusively in the nucleus, each arguing against
    the other by name. The edge from merlin loss to YAP activation is
    INDIRECT_KNOWN_INTERMEDIATES for that reason — the intermediates are known and
    their order is not.

Three phenotypes are deliberately not wired

The cataract, the retinal hamartoma and the neuropathy are left causally
unconnected. The reviews list them as manifestations and the criteria count them as
signs; no cited source says what produces them. An edge from the merlin node to a
lens opacity would assert a mechanism nothing here supports.

What is left out

The entry records that the second hit happens, and how often biallelic loss is found
in patients' schwannomas, but not the chromosomal mechanism by which the remaining
allele is lost. That is a gap, not an oversight, and a follow-up can close it.

Two scope questions, logged and not settled

Neither blocks this PR and neither touches a file this PR changes.

  • kb/disorders/Schwannomatosis.yaml binds MONDO:0008075, which MONDO's current
    release makes the parent of this entry's term — but that entry is written as the
    non-NF2 disease, with "non-NF2 schwannomatosis" among its synonyms and only
    SMARCB1- and LZTR1-related subtypes. After MONDO's move the entry bound to the
    parent class explicitly excludes what MONDO files as its child. MONDO is also
    internally inconsistent here: MONDO:0008075's own definition excludes vestibular
    involvement while the class is the asserted parent of a disease whose defining
    feature is bilateral vestibular schwannoma. The fix is that curator's to choose,
    and may belong upstream.
  • Neurofibromatosis.yaml carries an NF2 subtype row bound to this entry's term
    whose description says the entity "is currently uncurated as a standalone dismech
    entry". That becomes false when this merges and needs a curated_in: pointer.

Verification

Two adversarial passes ran before this push, split by axis, each given the entry and
its sources but not the reasoning behind them. Between them they found five claims
every automated gate accepted. Four were statements that something does not exist,
refuted by the ontologies or by a source the entry itself cites:

The entry said Actually
HPO has no ependymoma term HP:0002888 Ependymoma, two levels below the term bound
HPO has no bilateral vestibular schwannoma term HP:0009589, a child of the term bound
NCIT's clinical-action tree has no term for deciding not to act NCIT:C25598 Observation Activity is inside it, and Acoustic_Neuroma already uses it
No ClinGen assertion for this pair is cached ClinGen graded it Definitive in 2019, against this entry's own MONDO term

The fifth said no cited source measures how often the second hit occurs; a cited
paper puts biallelic NF2 loss at about 60% of patients' schwannomas. All five are
repaired and the three ontology terms rebound.

Also corrected: a prevalence band an order of magnitude out against its own rate; a
VERY_FREQUENT band on hearing impairment that no cited source supports (removed,
for the same reason the tinnitus row carries none); a stated source count of
thirty-seven against an actual nineteen; two mouse results graded DIRECT for an
edge neither quote names; a protocol paper's own data graded REVIEW_SYNTHESIS; a
result resting on in vivo Drosophila graded IN_VITRO; a preferred_term broader
than the term it was bound to; and three explanations claiming more than their
quotes said.

The GeneReviews chapter check-genereviews reported as UNTAGGED_CHAPTER is now
cited for the phenotype baseline and tagged in references:.

Validation

  • just validate — schema, terms and references clean; 93 snippets and 94 titles verified,
    0 issues
  • All 25 offline gates pass, including check-gene-activity-grounding, which this
    entry failed until both NF2-bearing nodes bound merlin's signalling adaptor
    activity
  • just check-genereviews --strict clean for GeneReviews
  • just list-disconnected-phenotypes — 7 of 10 connected
  • yamlfix is a no-op on the file

🤖 Generated with Claude Code

@cbhindex cbhindex self-assigned this Nov 13, 2026
@ai4c-reviewer

ai4c-reviewer Bot commented Nov 13, 2026 •

Copy link
Copy Markdown

Claude encountered an error after 0s —— View job


I'll analyze this and get back to you.

Curates the syndrome rather than its tumours: the germline lesion, the two-hit
mechanism, the 2022 diagnostic criteria, the tumour spectrum as phenotypes,
mosaicism as its own inheritance block, surveillance and outcome. The
constituent tumours stay in Acoustic_Neuroma, Meningioma, Schwannoma and
Ependymoma.

Six pathophysiology nodes, 21 pathograph nodes and 17 edges with no orphan
targets, 7 of 10 phenotypes wired, 93 verified snippets over 20 publications
plus the ClinGen Definitive gene-disease validity record for NF2.

Three discussions, one of them a CONTROVERSY: two full-text papers disagree
about whether merlin acts on LATS at the plasma membrane or in the nucleus, and
each argues against the other by name, so the edge from merlin loss to YAP
activation is INDIRECT_KNOWN_INTERMEDIATES.

Closes #17397

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@cbhindex
cbhindex force-pushed the nbp-nf2-schwannomatosis branch from 2e031fd to f9a1068 Compare November 16, 2026 13:38
@ai4c-reviewer

ai4c-reviewer Bot commented Nov 16, 2026 •

Copy link
Copy Markdown

Claude finished @cbhindex's task in 1m 48s —— View job


Review of PR #17478: changes requested

I submitted a REQUEST_CHANGES review. Overall this entry is strong: all 93 snippets verify against the cached references, the 25 changed files all belong to this entry, the pathophysiology nodes are atomic, and the ontology terms are good. Two things block it.

  • Checked the scope of the changed files
  • Read the entry
  • Spot-checked snippets, titles and terms
  • Ran the GeneReviews check
  • Submitted the review

🟡 Blocking

  1. GeneReviews is only partly used. The chapter is cited for clinical features and inheritance, but nothing is taken from its Diagnosis, Management or Surveillance sections. The most important gap: the chapter warns against radiation therapy for NF2-associated tumours, especially in childhood, because of the malignancy risk, and the stereotactic radiosurgery row says nothing about that. Three smaller gaps:
    • The diagnostic-criteria row paraphrases the chapter but cites other papers.
    • The MRI interval comes from a review of sporadic vestibular schwannoma, while the chapter gives the syndrome's own schedule (annual MRI from about age 10–12, annual hearing tests including BAER, annual eye exams).
    • The chapter's 50% versus less-than-50% transmission-risk sentence for mosaic parents is not quoted.
  2. Neurofibromatosis.yaml will be wrong once this merges. Its NF2 subtype row says the disease "is currently uncurated as a standalone dismech entry". Fixing it takes one line (curated_in: NF2-Related_Schwannomatosis), so it should go in this PR.
  3. No deep-research report. The review skill expects one for a new entry. Either run one, or have a maintainer explicitly waive the requirement.

🔵 Suggestions

  • Add hearing rehabilitation, especially auditory brainstem implants.
  • The YAP/TAZ node is bound to the same Hippo term as the node above it, so the two look identical by term.
  • Drop the counts in notes: that describe repository state, since they will go out of date.

@ai4c-reviewer ai4c-reviewer Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Review: NF2-Related Schwannomatosis

This is a well-built entry. The six-node chain is atomic, the membrane-vs-nucleus controversy is handled honestly, and the frequency bands are well calibrated. It is still blocked on two things. First, the GeneReviews chapter is only partly mined: nothing from its Diagnosis, Management or Surveillance sections is used, and Management includes a radiation warning that the entry contradicts by omission. Second, the PR leaves a statement in Neurofibromatosis.yaml that will be false once this merges.

  • Schema validation would pass (author's just validate; scope is clean: 25 files, all belonging to this entry)
  • All PMIDs verified as real (every cited reference is cached; titles spot-checked against cache)
  • Evidence snippets are exact quotes (count-verified-snippets: 93/93)
  • Ontology terms correct and specific (HP:0009589, HP:0002888, NCIT:C25598, CL triple on the second-hit node all look right)
  • evidence_source classifications correct
  • No pathophysiology bundling

🟡 IMPORTANT

1. GeneReviews is mined only for Clinical Characteristics and inheritance (review rubric item 15). PMID:20301380 is tagged and cited 6 times. None of those citations come from its Diagnosis or Management sections:

  • Agents/circumstances to avoid. GR says: "Radiation therapy of NF2-associated tumors, especially in childhood, when malignancy risks are likely to be substantially higher." The Stereotactic radiosurgery row says nothing about malignancy risk or avoiding radiation in children. Clinically, this is the most important gap. Add a GR evidence item and a caveat on that treatment.
  • Diagnosis/Testing. The Diagnostic criteria, 2022 revision description paraphrases GR almost word for word ("bilateral vestibular schwannomas, an identical NF2 pathogenic variant identified in two or more anatomically distinct NF2-related tumors…") but cites other papers for it. Cite GR for that sentence.
  • Surveillance. The MRI row takes its interval from a review of sporadic vestibular schwannoma (PMID:32232723, graded INDIRECT for that reason). GR gives the syndrome's own surveillance schedule: "annual brain MRI beginning at approximately age ten to 12 years…; annual hearing evaluation, including BAER testing; annual complete ophthalmology examination." Use it. A surveillance/monitoring treatment row would hold it naturally.
  • Genetic counseling. The mosaic block says transmission risk is "below the dominant one", but GR's explicit statement is not quoted: "offspring of an individual with a germline pathogenic variant have a 50% chance of inheriting the variant, while offspring of an individual who has mosaic NF2 may have a less than 50% risk".

2. Neurofibromatosis.yaml NF2 subtype row (line ~150) becomes false on merge. It says the entity "is currently uncurated as a standalone dismech entry". The fix is one line: add curated_in: NF2-Related_Schwannomatosis and trim that sentence. Please do it in this PR. Deferring it guarantees a window where the KB contradicts itself, and the repo's strictness policy asks for fixable problems not to be pushed to a follow-up. (The Schwannomatosis.yaml MONDO:0008075 scope question is different: it genuinely needs another curator's or MONDO's decision, and logging it is the right call.)

3. No deep-research artifact. The review skill expects at least one research/*-deep-research-*.md for a new entry. The entry is clearly thorough, so this is a process point. Either run one (and check it against the entry), or have a maintainer waive the requirement explicitly.

🔵 SUGGESTIONS

  • Hearing rehabilitation. GR lists audiology referral, cochlear and auditory brainstem implants, and lipreading/sign language. ABI is a defining NF2-specific intervention and is absent. Also consider GR's bevacizumab-for-ependymoma line, and its "average age of onset is 18 to 24 years" as an onset qualifier.
  • YAP and TAZ Activation binds the same hippo signaling DECREASED as its upstream node, so the two nodes look identical by term. Consider binding the downstream effect instead, e.g. a TEAD/YAP-dependent transcription process if GO has a fitting term. Otherwise leave this node's process unbound and let the description carry it.
  • Repository-state prose in notes: will go stale. Examples: "Twenty publications are cited… Nine of the twenty are cached as full text", and "The local build read on 21 October 2026…". The MONDO-parent fact is useful; the counts belong in the history record. Likewise, the "Written from cached sources only, by hand…" opening of review_notes describes the session, not the disease.
  • StatPearls Neurofibromatosis Type 2 exists (PMID:29261934). It is optional and never a baseline.

What's good

  • The three unwired phenotypes (cataract, retinal hamartoma, neuropathy) have a sound stated rationale.
  • REFUTE items in the CONTROVERSY and HUMAN_MODEL_MISMATCH discussions are used correctly.
  • Prevalence arithmetic checks out: 1/50,500 ≈ 2.0 and 1/27,500 ≈ 3.6 per 100k, with correct bands and denominators.
  • The GERMLINE_AND_SOMATIC rationale, the ClinGen Definitive record against the entry's own MONDO term, and the decision not to claim the unnamed 1-in-33,000 figure are all careful work.

@ai4c-reviewer ai4c-reviewer Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Re-review: NF2-Related Schwannomatosis. Changes are still needed, because nothing new has been pushed

The PR head is still f9a1068f50, the same commit the earlier changes-requested review looked at. I re-checked that commit on its own. The blocking items from that review are all still present, and I agree with each of them. The content is strong: the chain is atomic, the CONTROVERSY and model-mismatch discussions are handled well, the frequency bands and prevalence arithmetic are calibrated, and the ClinGen Definitive record is filed against the entry's own MONDO term. Below is what still has to change before this can merge.

  • Schema validation would pass (the author ran just validate; all 25 changed files belong to this entry, with no reversions or out-of-scope churn)
  • All PMIDs verified as real (every cited reference is cached)
  • Evidence snippets are exact quotes
  • Ontology terms correct and specific
  • evidence_source classifications correct
  • No pathophysiology bundling

🟡 IMPORTANT (blocking)

1. Only part of the GeneReviews chapter is used. just check-genereviews reports TAGGED for PMID:20301380. All six citations of it come from the Clinical Characteristics and inheritance sections: autosomal dominant inheritance, de novo mosaicism, bilateral VS by age 30, the definition sentence, childhood presentation, and mononeuropathy. The cached abstract has three more sections that are never quoted:

  • Agents/circumstances to avoid. The chapter says to avoid "Radiation therapy of NF2-associated tumors, especially in childhood, when malignancy risks are likely to be substantially higher." The Stereotactic radiosurgery row (line ~1383) does not mention this. It is the most clinically important omission. Add the GR evidence item and a caveat to that row.
  • Surveillance. The chapter gives the syndrome's own schedule: annual neurologic exam, annual brain MRI from about age 10–12, annual hearing evaluation including BAER, and annual ophthalmology exam. Quote it. At the moment the MRI interval rests on a review of sporadic VS (PMID:32232723, graded INDIRECT).
  • Diagnosis/testing. The description in the 2022-criteria diagnosis row closely tracks the GR wording, but it cites other papers. Cite GR as well.
  • Genetic counseling. Quote GR's 50% versus "less than 50%" transmission-risk sentence for mosaic parents on the mosaic inheritance block. That block currently states the claim without a source.

2. kb/disorders/Neurofibromatosis.yaml line 150 will become false once this merges. It still says NF2 is "currently uncurated as a standalone dismech entry". Add curated_in: NF2-Related_Schwannomatosis to that subtype row and trim the sentence. This is a one-line change and belongs in this PR.

3. No deep-research artifact. A new entry needs one. research/Schwannomatosis-deep-research-falcon.md is about the non-NF2 disease and does not count. Either run a report for this entry and cross-check it against the YAML, or get a maintainer to waive the requirement explicitly on this PR.

🔵 SUGGESTIONS

  • Add hearing rehabilitation, especially the auditory brainstem implant, which GR lists and which is specific to NF2. GR also gives an average onset age of 18–24 years, which you could use as an onset qualifier.
  • YAP and TAZ Activation binds the same hippo signaling DECREASED term as its upstream node. Bind the downstream transcriptional effect instead, or leave the process unbound.
  • Some prose describes repository or session state ("Twenty publications are cited…", "The local build read on…", and the opening of review_notes). Move it to the history record.
  • A StatPearls chapter exists (PMID:29261934). Citing it is optional, and it is never a baseline.

Please address everything in a single push. The approval gate clears when a re-review approves the new head.

The review's blocking finding was that the chapter was used only for Clinical
Characteristics and inheritance, and that its Management section carries a
radiation warning the entry contradicted by omission.

Four citations from the sections that were skipped. The agents-to-avoid
statement - radiation of NF2-associated tumours, especially in childhood, where
malignancy risk is judged substantially higher - is now a REFUTE item on the
stereotactic radiosurgery row, with the caveat in its description. The
chapter's own statement of the three diagnostic routes goes on the criteria
row, which had paraphrased it while citing other papers. The surveillance
schedule gets a treatment row of its own, because it follows the syndrome
rather than one known tumour. And the explicit transmission figures go on the
mosaic inheritance block.

A hearing-rehabilitation row is added: auditory brainstem implantation is the
option this syndrome needs and most causes of deafness do not, because the
tumours and their surgery destroy the cochlear nerve a cochlear implant
requires intact.

Neurofibromatosis.yaml said its NF2 subtype is uncurated as a standalone entry,
which this PR makes false. It now carries curated_in and corrected prose, here
rather than in a follow-up, so no window exists in which the KB contradicts
itself.

Two suggestions taken. The YAP and TAZ node bound the same hippo signaling
DECREASED as the node above it; it now binds its own effect. And the source
census and the dated MONDO-build mechanics are out of notes and into the
history record, since both described the repository rather than the disease.

NCIT has no surveillance term inside the clinical-intervention tree: C15719,
C61256, C48261 and C53607 all exist and none is reachable from C25218, so the
row binds Observation Activity and the note records the searches.

98 snippets verified, 19 offline gates green.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@cbhindex

Copy link
Copy Markdown
Contributor Author

Both blocking items are done, plus two of the suggestions. One item I can't satisfy myself and need a maintainer's call on — it's the last section below.

1. GeneReviews mined beyond Clinical Characteristics

You were right that the chapter was used for one section and the most clinically important thing in it was missing.

  • Agents to avoid. The radiation statement is now a REFUTE evidence item on the Stereotactic radiosurgery row, and its description says the row is not a first option in a child. Grading it REFUTE rather than SUPPORT is deliberate: it is the one cited sentence that argues against using the treatment, and burying that in a notes line would have left the same gap in a quieter form.
  • Diagnosis. The criteria row now cites the chapter's own statement of the three routes, which its description had been paraphrasing while citing the consensus paper for it.
  • Surveillance. A Surveillance of affected and at-risk individuals row carries the schedule. I made it a separate row rather than folding it into Observation with serial imaging, because the two answer different questions — one follows the syndrome and its at-risk relatives, the other follows a tumour already found — and the existing row's interval comes from a review of sporadic disease.
  • Genetic counselling. The mosaic block now quotes the 50% / less-than-50% statement instead of describing it as "below the dominant one".

2. Hearing rehabilitation

Added, and you were right that it is NF2-specific rather than generic: the brainstem implant is needed here precisely because bilateral vestibular schwannomas and their surgery destroy the cochlear nerve that a cochlear implant requires intact. The row states the options without outcomes, because the chapter lists availability and does not rank them.

A note on its binding, since it is the kind of thing worth recording: NCIT has no term for auditory brainstem implantation, and the device terms are not clinical actions so cannot sit in a TreatmentTerm. The action binds NCIT:C15315 Rehabilitation and the preferred term carries the specificity.

The same problem bit the surveillance row harder. NCIT:C15719 Surveillance fails the TreatmentActionTerm enum — it is not reachable from NCIT:C25218. Nor are NCIT:C61256 Monitoring, NCIT:C48261 Screening or NCIT:C53607 Patient Monitoring; I checked all four. The row binds NCIT:C25598 Observation Activity, which it now shares with the observation row, and the note records those four CURIEs so the next curator does not repeat the search.

3. Neurofibromatosis.yaml

Done in this PR, as you asked. curated_in: NF2-Related_Schwannomatosis plus corrected prose. Your reasoning is right — deferring it guarantees a window where the KB contradicts itself, and it is one line.

4. Suggestions taken

  • YAP and TAZ node. Good catch: it bound the same hippo signaling DECREASED as the node above it, so the two were indistinguishable by term. It now binds its own effect, GO:0045944 INCREASED, with the TEAD specificity in preferred_term. GO:0140552 TEAD-YAP complex exists but is a cellular component, so it cannot go in biological_processes.
  • Repository-state prose. Removed from notes: and review_notes:. The source census and the dated account of which MONDO build reports which parent were describing the repository, not the disease — the census is in the history record now, and the MONDO fact is reduced to the reclassification itself, which is durable. This is worth admitting: I added those counts in the previous round because a check found the old count wrong. Correcting a number in place was the wrong fix; the right one was to move it out of notes: entirely.

5. The deep-research artifact — needs a maintainer's decision

I cannot satisfy this one honestly. This entry was curated by hand from cached references, which is how this contributor's entries are built; there is no provider run behind it to commit. I am not going to generate a report after the fact and present it as the input the entry came from — that would misrepresent the provenance, and research/ is specifically for raw provider outputs.

So: either a maintainer waives the requirement for this entry, or it needs a real run before merge. Flagging rather than deciding. @cmungall — happy to be told which.

Validation

98 snippets verified on this entry, 50 on Neurofibromatosis.yaml, 0 issues. 19 offline gates pass, including terms and check-duplicate-keys. 7 of 10 phenotypes connected, unchanged.

🤖 Generated with Claude Code

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Curate NF2-related schwannomatosis (MONDO:0007039)

1 participant