Skip to content

Mondo-anchoring audit (dismech → Mondo)

Date: 2026-07-30 Issue: #7175 — tripartite gap-exchange (dismech ⇄ Monarch KG ⇄ Mondo) Scope: kb/disorders/*.yaml (n = 1,640). Fully local, deterministic. Regenerate:

uv run python scripts/mondo_anchor_audit.py                                   # summary
uv run python scripts/mondo_anchor_audit.py --tsv research/mondo_anchoring_worklist.tsv

This is the first of the six directed gap-flows in #7175 — the one that needs no external data. It measures how well dismech disorder records are anchored to Mondo, and surfaces the concrete dismech → Mondo handoff items (new-term / subtype requests) plus dismech-internal fixes. It does not decide how any individual concept should be modeled — that per-entry policy is #7178.

Headline

Metric Value
Disorders with a MONDO primary disease_term 1,611 / 1,640 (98.2%)
MONDO ids missing from the current Mondo release 0
MONDO ids obsolete/deprecated in Mondo 0
Stored-label drift vs Mondo canonical 3
Unanchored primary slot, no MONDO anywhere (Mondo candidates) 17
Unanchored primary slot, MONDO present elsewhere (promote-anchor fix) 12
mondo_mappings narrowMatch / broadMatch (granularity mismatch) 5 / 7
MONDO ids used as primary anchor by >1 entry 22 ids / 54 entries

The anchoring layer is in good shape: no dangling or obsolete Mondo ids. The actionable work splits cleanly into dismech → Mondo requests (17 concepts) and dismech-internal fixes (3 label drifts + 12 anchor promotions).

A. dismech → Mondo: new-term / subtype candidates (17)

Entries with no MONDO id anywhere in the file. Six carry an OMIM id that anchors a Mondo lookup or new-term request; the remainder are exposure / iatrogenic / viral / gene-level concepts that test Mondo's scope.

Entry OMIM anchor Note
EEFSEC_Deficiency OMIM:607695 gene-level neurodevelopmental
GOLGA2-Related_Golgin_A2_Deficiency OMIM:620240 gene-level
HAO1-Related_Glycolate_Oxidase_Deficiency OMIM:605023 gene-level metabolic
RAB5C-Related_Neurodevelopmental_Disorder_with_Macrocephaly OMIM:604037 gene-level
SLC26A6-Related_Hyperoxaluria_and_Nephrolithiasis OMIM:610068 gene-level transporter
VPS51-Related_Pontocerebellar_Hypoplasia-CDG OMIM:618606 gene-level CDG
KATNB1-related_Cortical_Malformation — gene-level
NDE1-related_Microcephaly_Lissencephaly — gene-level
TUBB_TUBB5-related_Microcephaly — gene-level
UGGT1-congenital_disorder_of_glycosylation — gene-level CDG (has term, no id)
Arsenic_Poisoning — exposure/toxicology
Chemotherapy_Induced_Nausea_and_Vomiting — iatrogenic
Spaceflight_Associated_Neuro-Ocular_Syndrome — exposure/environmental
Volumetric_Muscle_Loss — acquired injury
Transient_Neonatal_Pustular_Melanosis — dermatologic (has term, no id)
Human_Metapneumovirus_Infection — infectious
Seasonal_Coronavirus_Infection — infectious

B. dismech-internal fixes (no Mondo action)

B1. Label drift — stored term.label ≠ Mondo canonical (3)

Entry MONDO id Stored label Canonical label
Chronic_Myeloid_Leukemia MONDO:0011996 chronic myelogenous leukemia, BCR-ABL1 positive chronic myeloid leukemia
Minimal_Change_Disease MONDO:0006835 lipoid nephrosis minimal change disease
Multiple_Mitochondrial_Dysfunctions_Syndrome_9B MONDO:0971174 multiple mitochondrial dysfunctions syndrome 9B multiple mitochondrial dysfunctions syndrome 9b

The first two store a Mondo synonym rather than the canonical label; the third is a case-only difference. preferred_term may stay as-is (it is allowed to differ); term.label should be corrected to the canonical string.

B2. Promote-anchor — MONDO present elsewhere but not in the primary slot (12)

These already reference a plausible MONDO id (in mappings/genetic) but leave disease_term.term.id empty. The fix is to promote the correct id into the primary slot — no Mondo request needed. Care is required where several MONDO ids appear (some are comorbidities/related terms, not the entity's own class):

AIP-related_pituitary_adenoma_predisposition, Acute_Post-Surgical_Pain, Adenovirus_Respiratory_Infection, CKD-Mineral_Bone_Disorder, FICUS_syndrome, GNAS-related_pituitary_adenoma_3, GPR101-related_pituitary_adenoma_2, Green_Tobacco_Sickness, MCM9-related_gametogenic_failure, SLC26A1-Related_Oxalate_Transporter_Deficiency, SRPX2-related_Speech_Epilepsy_Polymicrogyria, USP8-related_pituitary_adenoma_4.

The three pituitary-adenoma entries all point at MONDO:0006373 (pituitary adenoma) — a signal they may want a shared intermediate anchor with gene-specific subtypes, which is exactly the record-altitude question in #7178.

C. Granularity mismatch (mondo_mappings)

These are candidate inputs to the mechanism-gated split decision (#7178), recorded here only as discrepancies — not adjudicated.

  • narrowMatch (dismech finer than mapped Mondo), 5: ER_Positive_Breast_Cancer, Familial_Thoracic_Aortic_Aneurysm_and_Aortic_Dissection (×9), GABRG2-Related_Epilepsy, GRIN1-Related_Neurodevelopmental_Disorder (×3), Parkinsons_Disease.
  • broadMatch (dismech broader than mapped Mondo), 7: Acetaminophen_Hepatotoxicity, Focal_Articular_Cartilage_Defect_of_the_Knee, GABRG2-Related_Epilepsy, Rhizomelic_Chondrodysplasia_Punctata_Plasmalogen_Synthesis_Defect (×2), Stargardt_Disease, Stickler_Syndrome_Type_1, Trimethylaminuria.

mondo_mappings skos-predicate totals: exactMatch 288 · closeMatch 30 · narrowMatch 15 · relatedMatch 14 · broadMatch 8. (Predicate occurrences; the narrow/broad file counts above are distinct entries.)

D. Shared primary anchors (same MONDO id on >1 entry)

22 MONDO ids are used as the primary disease_term by 54 entries. Detected by scripts/mondo_anchor_audit.py; per-entry counts are in the shared_anchor_n column of the worklist TSV. Three tiers:

Tier 1 — intentional finer-than-Mondo splits (~45 entries; correct as-is)

Precision-oncology driver subtypes and metastatic-stage variants sharing a parent term because Mondo has no molecular-subtype term. Simultaneously a strong dismech → Mondo subtype-request signal.

MONDO id Label × Entries
MONDO:0005061 lung adenocarcinoma 5 EGFR / KRAS-G12C / MET-ex14 / RET / ROS1 NSCLC
MONDO:0005233 non-small cell lung carcinoma 4 ALK / BRAF-V600E / Metastatic / generic
MONDO:0005575 colorectal cancer 4 BRAF-V600E / HER2+ / MSI-high / Metastatic
MONDO:0005012 cutaneous melanoma 3 BRAF-V600 / Metastatic / NRAS
MONDO:0005075 thyroid gland papillary carcinoma 3 BRAF / generic / RET-fusion
MONDO:0007256 hepatocellular carcinoma 3 Aflatoxin / generic / Metastatic
MONDO:0001056 gastric cancer 2 EBV / HER2+
MONDO:0003210 intrahepatic cholangiocarcinoma 2 FGFR-altered / IDH-mutant
MONDO:0004950 gastric carcinoma 2 H. pylori / Metastatic
MONDO:0004989 breast carcinoma 2 Metastatic / PIK3CA
MONDO:0005086 renal cell carcinoma 2 Metastatic / generic
MONDO:0005211 ovarian serous adenocarcinoma 2 Metastatic / HGSC
MONDO:0008315 prostate cancer 2 BRCA / Metastatic
MONDO:0007915 systemic lupus erythematosus 2 Neuropsychiatric SLE / generic
MONDO:0100038 complex neurodevelopmental disorder 2 ANK2 / BLOC1S1 (two-gene-at-generic-term)

Tier 2 — likely mis-anchor (a more specific Mondo term probably exists)

  • Confirmed: the Waardenburg pair both sit on generic MONDO:0018094, but type-level terms exist — MONDO:0008670 (WS type 1) for PAX3_Waardenburg_Spectrum, MONDO:0019517 (WS type 2) for MITF_Waardenburg_Tietz_Spectrum. Caveat: these are "spectrum" records and may be intentionally broader than one numbered type — curator's call.
  • Worth a targeted check (OAK search inconclusive, not asserting absence): Obesity_Due_to_MC4R_Pathway_Disruption (on MONDO:0011122 obesity disorder); SLC6A1-Related_Disorder (on MONDO:0014633 epilepsy with myoclonic-atonic seizures, likely too narrow for the gene's full phenotype); Malnutrition-related_Diabetes_Mellitus (on the diabetes umbrella MONDO:0005015, which the umbrella entry itself already holds via closeMatch); Pacak-Zhuang_syndrome (on MONDO:0035540 pheochromocytoma-paraganglioma).

Tier 3 — possible genuine redundancy / lump

  • Neuromyelitis_Optica + Neuromyelitis_Optica_Spectrum_Disorder → MONDO:0019100. Mondo folds NMOSD into the same term (no separate NMOSD class found), so these may be the same entity modeled twice — the clearest duplication candidate.
  • Chemotherapy_Induced_Diarrhea + Travelers_Diarrhea → MONDO:0001673 (diarrheal disease). Two etiologically unrelated conditions pinned to a generic symptom-level term; both are under-anchored.

Tiers 2/3 are flagged for curator review, not auto-fixed — they intersect the record-altitude policy call (#7178).

E. Groupings & modules anchoring

Different classes anchor differently, so the disorder audit doesn't apply verbatim. Regenerate: uv run python scripts/grouping_anchor_audit.py.

Groupings (kb/groupings/, n=49). A Grouping carries an optional MONDO cross-ref in mappings.mondo_mappings — not a disease_term (a grouping stands on its own curated rationale and need not recapitulate a MONDO class).

  • 30 map to MONDO — all valid (0 missing / obsolete / label-drift).
  • 19 have no MONDO mapping. Two sub-cases (OAK basic_search is imperfect, so treat as provisional):
  • Existing MONDO term → just add the mapping: Mucolipidoses (MONDO:0019248).
  • Likely novel mechanism/treatment-response unions → Mondo grouping-class candidates: Fibrotic Disorders, Polyglutamine Disorders, TDP-43 Proteinopathies, FGFR-Related Skeletal Dysplasias, DNA Repair Synthetic-Lethality Cancers, Immune Checkpoint-Responsive Cancers, Digenic and Oligogenic Disorders, Parkinsonism Dopaminergic Degeneration Disorders, Epilepsy Excitation-Inhibition Imbalance Disorders, and others (full list in the script output). Many are mechanism-based unions Mondo is unlikely to carry — consistent with dismech leading Mondo on mechanism-defined groupings.

Modules (kb/modules/, n=118). Deliberately not Mondo-anchored — they model conserved processes anchored to process ontologies (GO / OGMS / MPATH / UBERON), so they are out of scope for the Mondo dimension. 0 carry a disease_term; the 11 with a stray MONDO reference (in evidence/notes) carry no obsolete ids. (The MPATH continuant gaps for Xogenesis modules — amyloid/thrombus/atheroma — are a separate OBO-request track.)

Corrections to earlier ad-hoc figures

An earlier chat-level count (before this reproducible pass) was wrong on three points, recorded here so the numbers don't propagate:

  • "17 unanchored" → 29 primary-slot (17 no-MONDO-anywhere + 12 promote-anchor).
  • "9 narrowMatch subtype candidates" → 5 MONDO-narrower entries (the earlier count swept in ICD10CM/NCIT narrowMatches from other *_mappings blocks).
  • "6 files with explicit gap flags" → a crude text-regex detector was unreliable and is omitted; explicit MONDO lacks … prose flags need a better pass before being trusted.
  • Groupings & modules anchoring — §E above (grouping_anchor_audit.py).
  • Mondo → dismech (curation-target direction) — docs/reports/mondo-to-dismech-gaps-2026-07-31.md (mondo_to_dismech_gaps.py), bounded to the neighborhood of curated terms.
  • Monarch KG ⇄ dismech — gene–disease (kg-gene-gap-audit-2026-07-30.md) and disease–phenotype (kg-phenotype-gap-audit-2026-07-31.md).

Machine-readable worklist

Full per-disorder inventory (all 1,640 rows): research/mondo_anchoring_worklist.tsv (name, anchor_state, mondo_id, stored_label, oak_flag, mondo_mapping_preds, shared_anchor_n — the number of entries sharing that MONDO id as a primary anchor).