Skip to content

OMIM phenotypic series as dismech groupings — a heterogeneity audit

Date: 2026-08-01 Data: MONDO releases/2026-07-06 (mondo.obo), dismech kb/ at time of writing Script: scripts/omimps_grouping_audit.py Table: data/omimps-phenotypic-series-heterogeneity-2026-08-01.tsv

The question

A Mondo audit (2026-07-31) found 42 broad clinical grouping classes carrying has_characteristic MONDO:0021152 ! inherited, sourced solely to an OMIM phenotypic series (OMIMPS) equivalentTo xref. Two questions follow for dismech:

(a) Does it make sense to create kb/groupings/ entries for these OMIMPS-derived objects (or are some of them just distinct diseases)?

(b) How heterogeneous is each — in particular, does it combine genetic and acquired forms?

Short answer to both: an OMIM phenotypic series is not one kind of thing, and only one of the four kinds is a grouping. 16 of the 42 (38%) are not series of diseases at all — they are series of risk loci or linkage-mapped loci for a single disease, and the correct dismech treatment is one Disease entry with genetic: risk-factor rows, not a grouping with members. Celiac disease, the case that prompted the ticket, is the cleanest example.

Method

Everything below is computed offline from mondo.obo, so it is reproducible:

uv run python scripts/omimps_grouping_audit.py --tsv out.tsv     # downloads mondo.obo if needed
uv run python scripts/omimps_grouping_audit.py --detail MONDO:0005130
uv run python scripts/omimps_grouping_audit.py --focus all       # all 608 OMIMPS-equivalent classes

Each audited class's is-a descendants are typed:

Tier Signal in MONDO
MENDELIAN has_material_basis_in_germline_mutation_in <gene>
SUSCEPTIBILITY predisposition subset, or "susceptibility to" in the label (the OMIM {braces} convention)
UNMAPPED_LOCUS an OMIM: equivalentTo xref but no gene relation — the proxy for OMIM phenotype mapping key 2 (locus mapped, gene unknown)
SOMATIC has_material_basis_in_somatic_mutation_in
INFECTIOUS under MONDO:0005550, or disease_has_infectious_agent
ACQUIRED label names the acquiring process (senile, drug-induced, traumatic, diabetic, …)
UNSPECIFIED everything else — morphological/clinical subdivisions, groupers

Each descendant gets exactly one tier: the signals are not independent flags but are tested in precedence order INFECTIOUS > SUSCEPTIBILITY > MENDELIAN > SOMATIC > ACQUIRED > UNMAPPED_LOCUS. Two of those orderings carry the argument. Susceptibility outranks Mendelian, so an OMIM {braces} risk locus that MONDO has nonetheless given a gene relation is counted as a risk locus, not as a Mendelian disease — the distinction this whole audit turns on. And the gene relation outranks the ACQUIRED label test, so a gene-defined disease whose label happens to contain "diabetic" is not miscounted as acquired. The practical consequence for anyone summing columns out of the TSV: a class can appear in R while also carrying a gene.

Descendants alone would miss the most important members. MONDO deliberately keeps OMIM {susceptibility} entries out of the is-a tree of the disease they predispose to: they are is_a MONDO:0020573 inherited disease susceptibility, with an explicit excluded_subClassOf and a predisposes_towards link. So the count of inbound predisposers is reported alongside, and is the single sharpest signal that an OMIMPS is a risk-locus series.

Known limits. The ACQUIRED count is a floor, not an estimate — it can only see acquired forms MONDO actually models under the class, and MONDO models them sparsely (see "Latent vs contradicted" below). The UNMAPPED_LOCUS tier is a proxy: OMIM's phenotype mapping key is not in MONDO, so a locus whose gene MONDO simply has not linked yet is indistinguishable from a true linkage-only locus. The mondo_recipient_candidate column is advisory string-matching and needs a human read.

The celiac case, in full

$ uv run python scripts/omimps_grouping_audit.py --detail MONDO:0005130
# celiac disease (MONDO:0005130) OMIMPS:212750  kind=SUSCEPTIBILITY_SERIES
* UNSPECIFIED     MONDO:0800124    Lane Hamilton syndrome
  PREDISPOSES_TO  MONDO:0008930    celiac disease, susceptibility to, 1
  PREDISPOSES_TO  MONDO:0012340    celiac disease, susceptibility to, 2
  …                                (13 in total)

PS212750's 15 rows are all {Celiac disease, susceptibility to, N} — brace-annotated susceptibility entries, not diseases. Only three rows carry phenotype mapping key 3 (molecular basis known): the two 6p21.32 rows, HLA-DQA1 and HLA-DQB1, and CELIAC4 at 19p13.1. The remaining twelve are key 2 — a linkage/association interval with no gene: the eleven anonymous celiac loci (CELIAC2, 3, 5–13) plus the co-mapped 4q27 {Autoimmune disease, susceptibility to, 5} row. Several are recognisable as GWAS regions rather than loci in the Mendelian sense; CELIAC6 at 4q27, for instance, is the IL2/IL21 region.

Three consequences:

  1. There is nothing to group. MONDO has all thirteen susceptibility classes, and has already ruled that they are not subclasses of celiac disease — excluded_subClassOf MONDO:0005130 is asserted explicitly on them. The only is-a descendant of celiac disease in MONDO is Lane-Hamilton syndrome (celiac disease with idiopathic pulmonary hemosiderosis), which is not a member of the series. A kb/groupings/Celiac.yaml would have no members.
  2. The disease is genuinely gene–environment, not inherited. Celiac disease requires both an HLA-DQ2/DQ8 background and dietary gluten; monozygotic-twin concordance is incomplete and the disease remits on gluten withdrawal. It is not a hereditary disease in the sense MONDO:0003847 defines ("caused by genetic modifications … inherited from a parent's genome"), and dismech already types it correctly: category: Complex, inheritance: Polygenic inheritance (HP:0010982), with gluten exposure in the environmental: block.
  3. The dismech shape is already right — one Disease entry whose risk loci live in genetic: as risk-factor rows. This PR makes that shape explicit (see "Changes made").

The Mondo audit's disposition for celiac disease was STRIP, and this analysis agrees: the OMIMPS equivalentTo xref is the sole remaining source of the inherited claim, and MONDO's own excluded_subClassOf axioms already contradict the claim it produces.

Why it matters more than descendant counts suggest

The audit measured harm by descendant count, so classes with one or zero descendants (celiac disease, visceral leishmaniasis, IgA glomerulonephritis, thyrotoxic periodic paralysis) looked harmless. In the materialized release they are not. hereditary disease is only one of a family of classes defined as <genus> and has_characteristic some inherited, and a class picks up every one whose genus it also falls under. What that produces today:

Class is-a superclasses acquired from the inherited claim
Waldenstrom macroglobulinemia MONDO:0100280 hereditary disease; inborn errors of metabolism
thyrotoxic periodic paralysis MONDO:0019201 familial periodic paralysis; inborn errors of metabolism; hereditary neurological disease
primary biliary cholangitis MONDO:0005388 cirrhosis, familial
IgA glomerulonephritis MONDO:0005342 hereditary nephritis; inherited kidney disorder
age-related macular degeneration MONDO:0005150 inherited retinal dystrophy; hereditary neurological disease
otosclerosis MONDO:0005349 inherited auditory system disease
progressive supranuclear palsy MONDO:0019037 inherited neurodegenerative disorder
temporal lobe epilepsy MONDO:0005115 familial partial epilepsy
carpal tunnel syndrome MONDO:0007275 hereditary neuromuscular disease; hereditary neurological disease
systemic lupus erythematosus MONDO:0007915 hereditary disorder of connective tissue
psoriasis MONDO:0005083, chronic mucocutaneous candidiasis MONDO:0015279 hereditary skin disorder
visceral leishmaniasis MONDO:0005445 hereditary disease — for a parasitic infection

A B-cell lymphoma classified as an inborn error of metabolism, and sporadic PSP classified as an inherited neurodegenerative disorder, are not latent problems. The damage is lateral (each class gains spurious hereditary-X superclasses) as well as downward, and it is independent of how many descendants the class has.

One false positive. MONDO:0020836 autism, susceptiblity to is correctly placed: it is already is_a MONDO:0020573 inherited disease susceptibility, is in the predisposition subset, and all 26 of its descendants are susceptibility classes. The inherited characteristic is appropriate on a susceptibility grouper. It should come off the audit's PATTERN list.

Latent vs contradicted: the acquired axis is under-counted

The audit's "contradicted today" column counts descendants already provably non-genetic. It reads low (7 across all 42) because MONDO models acquired forms of these entities sparsely, not because the entities lack them. Two illustrations from the same release:

  • Diabetic cataract (MONDO:0001687) and diabetes-mellitus-type-2-associated cataract (MONDO:0005408) are children of cataract, so they inherit the hereditary claim today. The Mondo audit's "contradicted" column did not count them because their labels carry no "senile"/"acquired" marker; this audit initially missed them for the same reason, and counts them only because the ACQUIRED label list was extended with diabet(ic|es) after review. That is the shape of the whole problem: the count moves with the vocabulary you happen to have enumerated.
  • Drug-induced lupus erythematosus (MONDO:0016474) escapes only by an accident of placement: it is a sibling of SLE under lupus erythematosus, not a child.
  • Tetanic cataract (MONDO:0001811, hypocalcaemic) is still uncounted here — an acquired form whose label names neither an agent nor an age.

Neither MONDO nor this audit models secondary craniosynostosis, acquired hypogonadotropic hypogonadism, secondary parkinsonism, or acquired bronchiectasis at all — every one a real, common clinical entity that would silently inherit the defect the moment someone adds it. So read the A column as "acquired forms MONDO happens to have", and the MENDELIAN+RISK_LOCUS axis label as "no acquired form in MONDO", not "no acquired form".

The four kinds of OMIM phenotypic series

Kind n What the members are dismech shape
SUSCEPTIBILITY_SERIES 7 Brace-annotated OMIM risk loci; MONDO holds them outside the is-a tree as predisposers one Disease, loci as genetic: rows with relationship_type: SUSCEPTIBILITY
LOCUS_SERIES 9 Numbered linkage intervals, gene mostly unidentified (MYP1-25, IBD1-30, PBC1-5, OTSC1-10) one Disease; add members only as genes are identified
MIXED_GENETIC_ACQUIRED 3 Gene-defined diseases and infectious/acquired forms under one clinical umbrella grouping over the genetic subset only — never over the clinical parent
MENDELIAN_SERIES 22 Germline gene-defined diseases; no acquired form in MONDO grouping is well-formed, if dismech has ≥3 member entries

Heterogeneity axes across all 42: M+R 24, R only 8, M only 6, M+R+A 3, M+A 1. So 28 of 42 (67%) mix at least two member kinds, and only 6 are a clean Mendelian series (a further 8 are pure risk-locus series with no Mendelian member at all). The dominant mixture is not genetic-plus-acquired — it is genetic-plus-risk-locus, the OMIM series bundling {susceptibility} entries with gene-defined Mendelian diseases under one PS number. That is the heterogeneity that actually bites, and it is invisible in the inherited flag because both kinds of member are "genetic" in a loose sense.

Beyond the 42

--focus all runs the same typing over all 608 MONDO classes carrying an OMIMPS equivalentTo xref, not just the 42 broad ones the audit flagged:

Series kind n %
MENDELIAN_SERIES 457 75%
LOCUS_SERIES 49 8%
SUSCEPTIBILITY_SERIES 31 5%
SPARSE 62 10%
MIXED_GENETIC_ACQUIRED 9 1%

So the OMIMPS pattern is right three times out of four, and the risk-locus problem is a 13% tail (80 classes) rather than a systematic failure — but that tail is heavily concentrated in exactly the broad clinical classes the audit selected (38% there vs 13% overall). dismech already anchors a Disease entry on 231 of the 608, which is the natural worklist for a follow-up sweep.

Full table

M / R / A = Mendelian / risk-locus (susceptibility + unmapped locus + predisposers) / acquired (infectious + acquired + somatic) member counts.

Class OMIMPS Series kind M / R / A Axes dismech members dismech disposition
autism, susceptiblity to MONDO:0020836 PS209850 SUSCEPTIBILITY_SERIES 0 / 26 / 0 R 0 SINGLE_DISEASE
systemic lupus erythematosus MONDO:0007915 PS601744 SUSCEPTIBILITY_SERIES 8 / 19 / 0 M+R 0 SINGLE_DISEASE
psoriasis MONDO:0005083 PS177900 SUSCEPTIBILITY_SERIES 2 / 12 / 0 M+R 1 SINGLE_DISEASE
celiac disease MONDO:0005130 PS212750 SUSCEPTIBILITY_SERIES 0 / 13 / 0 R 0 SINGLE_DISEASE
thyrotoxic periodic paralysis MONDO:0019201 PS188580 SUSCEPTIBILITY_SERIES 0 / 3 / 0 R 0 SINGLE_DISEASE
visceral leishmaniasis MONDO:0005445 PS608207 SUSCEPTIBILITY_SERIES 0 / 3 / 0 R 0 SINGLE_DISEASE
Waldenstrom macroglobulinemia MONDO:0100280 PS153600 SUSCEPTIBILITY_SERIES 0 / 2 / 0 R 0 SINGLE_DISEASE
inflammatory bowel disease MONDO:0005265 PS266600 LOCUS_SERIES 12 / 22 / 3 M+R+A 0 SINGLE_DISEASE
myopia MONDO:0001384 PS160700 LOCUS_SERIES 10 / 19 / 0 M+R 0 SINGLE_DISEASE
orofacial cleft MONDO:0000358 PS119530 LOCUS_SERIES 9 / 13 / 0 M+R 1 SINGLE_DISEASE
chronic mucocutaneous candidiasis MONDO:0015279 PS114580 LOCUS_SERIES 5 / 7 / 0 M+R 1 SINGLE_DISEASE
keratoconus MONDO:0015486 PS148300 LOCUS_SERIES 2 / 7 / 0 M+R 0 SINGLE_DISEASE
otosclerosis MONDO:0005349 PS166800 LOCUS_SERIES 1 / 10 / 0 M+R 0 SINGLE_DISEASE
primary biliary cholangitis MONDO:0005388 PS109720 LOCUS_SERIES 0 / 5 / 0 R 0 SINGLE_DISEASE
preeclampsia MONDO:0005081 PS189800 LOCUS_SERIES 2 / 3 / 0 M+R 0 SINGLE_DISEASE
multinodular goiter MONDO:0000334 PS138800 LOCUS_SERIES 0 / 3 / 0 R 0 SINGLE_DISEASE
immunodeficiency disease MONDO:0021094 PS300755 MIXED_GENETIC_ACQUIRED 182 / 56 / 1 M+R+A 16 GROUPING_CANDIDATE
cataract MONDO:0005129 PS116200 MIXED_GENETIC_ACQUIRED 35 / 15 / 5 M+R+A 0 NO_DISMECH_BASIS
central precocious puberty MONDO:0019165 PS176400 MIXED_GENETIC_ACQUIRED 2 / 0 / 3 M+A 0 NO_DISMECH_BASIS
spermatogenic failure MONDO:0004983 PS258150 MENDELIAN_SERIES 106 / 7 / 0 M+R 0 NO_DISMECH_BASIS
craniosynostosis MONDO:0015469 PS123100 MENDELIAN_SERIES 34 / 27 / 0 M+R 6 COVERED_BY_GROUPING
hypogonadotropic hypogonadism MONDO:0018555 PS147950 MENDELIAN_SERIES 56 / 7 / 0 M+R 6 GROUPING_EXISTS
arthrogryposis multiplex congenita MONDO:0015168 PS617468 MENDELIAN_SERIES 26 / 5 / 0 M+R 3 GROUPING_CANDIDATE
Parkinson disease MONDO:0005180 PS168600 MENDELIAN_SERIES 14 / 12 / 0 M+R 3 GROUPING_CANDIDATE
holoprosencephaly MONDO:0016296 PS236100 MENDELIAN_SERIES 11 / 5 / 0 M+R 3 GROUPING_EXISTS
lymphoproliferative syndrome MONDO:0016537 PS308240 MENDELIAN_SERIES 11 / 2 / 0 M+R 3 COVERED_BY_GROUPING
hypotrichosis MONDO:0003037 PS605389 MENDELIAN_SERIES 13 / 4 / 0 M+R 1 GROUPING_DEFERRED
visceral heterotaxy MONDO:0018677 PS306955 MENDELIAN_SERIES 14 / 3 / 0 M+R 0 NO_DISMECH_BASIS
age-related macular degeneration MONDO:0005150 PS603075 MENDELIAN_SERIES 10 / 6 / 0 M+R 0 NO_DISMECH_BASIS
hyperinsulinemic hypoglycemia MONDO:0005803 PS256450 MENDELIAN_SERIES 8 / 0 / 0 M 0 NO_DISMECH_BASIS
paraganglioma MONDO:0000448 — MENDELIAN_SERIES 7 / 0 / 0 M 0 NO_DISMECH_BASIS
temporal lobe epilepsy MONDO:0005115 PS600512 MENDELIAN_SERIES 4 / 4 / 0 M+R 0 NO_DISMECH_BASIS
Moyamoya disease MONDO:0016820 PS252350 MENDELIAN_SERIES 4 / 4 / 0 M+R 0 NO_DISMECH_BASIS
hydatidiform mole MONDO:0006248 PS231090 MENDELIAN_SERIES 4 / 0 / 0 M 0 NO_DISMECH_BASIS
bone Paget disease MONDO:0005382 PS167250 MENDELIAN_SERIES 3 / 2 / 0 M+R 0 NO_DISMECH_BASIS
fetal and neonatal alloimmune thrombocytopenia MONDO:0019415 PS621264 MENDELIAN_SERIES 3 / 0 / 0 M 0 NO_DISMECH_BASIS
spastic quadriplegic cerebral palsy MONDO:0016215 PS612900 MENDELIAN_SERIES 3 / 0 / 0 M 0 NO_DISMECH_BASIS
progressive supranuclear palsy MONDO:0019037 PS601104 MENDELIAN_SERIES 2 / 2 / 0 M+R 0 NO_DISMECH_BASIS
bronchiectasis MONDO:0004822 PS211400 MENDELIAN_SERIES 2 / 1 / 0 M+R 0 NO_DISMECH_BASIS
chronic recurrent multifocal osteomyelitis MONDO:0009813 PS609628 MENDELIAN_SERIES 2 / 1 / 0 M+R 0 NO_DISMECH_BASIS
carpal tunnel syndrome MONDO:0007275 PS115430 MENDELIAN_SERIES 2 / 0 / 0 M 0 NO_DISMECH_BASIS
IgA glomerulonephritis MONDO:0005342 — SPARSE 0 / 1 / 0 R 0 NO_DISMECH_BASIS

dismech members counts distinct dismech Disease entries anchored on gene-defined descendants — not distinct MONDO ids, because one entry can anchor several ids through has_subtypes, and a one-entry "series" is a subtype catalog, not a grouping.

(a) Does it make sense to create dismech groupings?

For 16 of 42, no — and not because of curation capacity, but because the object is not a union of diseases. A dismech Grouping is an explicit, curated union that lists its members; where the OMIMPS members are risk loci that MONDO itself keeps out of the disease hierarchy, a grouping would be an empty union. These belong in an existing Disease entry's genetic: block, exactly as GeneDiseaseRelationshipEnum already provides for:

genetic:
- name: PTPN22
  gene_term: {preferred_term: PTPN22, term: {id: hgnc:9652, label: PTPN22}}
  association: GWAS
  relationship_type: SUSCEPTIBILITY   # "in combination with other genetic or environmental factors"

dismech already holds a Disease entry on 19 of the 42 audited classes (celiac disease, psoriasis, SLE, PBC, IgA nephropathy, Waldenström, preeclampsia, Moyamoya, PSP, …), so for most of the SINGLE_DISEASE rows there is nothing to create — only risk-locus rows to enrich.

For a further 18, not yet, for a different reason: these are genuine Mendelian series (spermatogenic failure with 106 gene-defined members, visceral heterotaxy with 14, …) but dismech curates none or one of their members. A grouping is a union over existing entries; these are curation targets first, groupings later.

For 5, yes — as groupings over the genetic subset, never over the clinical parent. Two of these are built in this PR (marked ✅ below). In descending order of readiness (member entries already in kb/disorders/ shown):

  1. immunodeficiency disease (16 member entries). But see below — dismech has already solved this one.
  2. ✅ hypogonadotropic hypogonadism (6): FGFR1-Related Hypogonadotropic Hypogonadism, Boucher-Neuhauser, Cerebellar Ataxia-Hypogonadism, Woodhouse-Sakati, Schaaf-Yang, Bosma Arhinia Microphthalmia. MONDO already has the recipient class, MONDO:0015770 congenital hypogonadotropic hypogonadism (32 gene-defined descendants). The grouping as built has 7 members: curating it added Kallmann Syndrome, which the automated count missed because MONDO models MONDO:0018800 as a grouper with no has_material_basis_in_germline_mutation_in relation, so it typed as UNSPECIFIED rather than MENDELIAN. Expect the automated member counts to run one or two low for this reason wherever the disease has a genetically heterogeneous grouper class.
  3. arthrogryposis multiplex congenita (3): Marden-Walker, Wieacker-Wolff, X-Linked Infantile SMA.
  4. Parkinson disease (3): PRKN-Related Juvenile PD, PARK7-Related Early-Onset PD, Kufor-Rakeb. Note dismech already has Parkinsonism_Dopaminergic_Degeneration_Disorders, whose members are Parkinson's Disease and Manganism — an idiopathic + acquired union. The monogenic entries are not members. Either extend it or mint a sibling Monogenic_Parkinson_Disease; do not create a grouping that silently re-lumps the acquired member.
  5. ✅ holoprosencephaly (3): SHH Holoprosencephaly Spectrum, HPE9, HPE12.

For 2, already covered. craniosynostosis members sit in FGFR-Related Skeletal Dysplasias; lymphoproliferative syndrome members sit in Immune Dysregulation IEIs. No new grouping.

dismech has independently already done the recommended fix for the largest class. The Mondo audit proposes minting a genetic-form child under hereditary.yaml for MONDO:0021094 immunodeficiency disease. dismech's Inborn_Errors_of_Immunity grouping is mapped skos:exactMatch to MONDO:0003778 inborn error of immunity — which already exists, and which is precisely the right recipient. The pathology is that in MONDO, MONDO:0003778 is a sibling of MONDO:0021094 (both under immune system disorder), and it is the clinical class, not the inborn-error class, that carries the inherited characteristic. AIDS being classified as a hereditary disease is the direct consequence. Moving OMIMPS:300755 and the characteristic onto MONDO:0003778 needs no new class.

(b) Genetic vs acquired heterogeneity

Only three of the 42 mix genetic and acquired members in MONDO today (immunodeficiency disease, cataract, central precocious puberty) — and, as flagged above, that is a floor set by MONDO's coverage of acquired forms, not a finding about the diseases. Of the audited classes, the ones whose real-world scope is unambiguously genetic+acquired and where the grouping boundary therefore has to be drawn carefully:

  • immunodeficiency disease — primary (inborn errors of immunity) vs secondary (HIV, drug-induced, malignancy-associated, malnutrition). MONDO already exhibits the contradiction via AIDS.
  • cataract — congenital/genetic vs senile, diabetic, traumatic, steroid-induced, radiation. Three senile classes are counted; diabetic cataract is a descendant and not counted by the label heuristic.
  • inflammatory bowel disease — monogenic VEO-IBD (IL10RA/IL10RB, XIAP, …) vs polygenic adult Crohn/UC vs infectious colitis, two of which sit under the class today. MONDO has no "monogenic IBD" grouper, so this one would need a new class minted.
  • craniosynostosis — syndromic/gene-defined vs secondary (metabolic, hematologic, deformational). MONDO models none of the secondary forms, but MONDO:0015338 syndromic craniosynostosis (30 gene-defined descendants) is a ready-made recipient.
  • hypogonadotropic hypogonadism — congenital (Kallmann and normosmic forms) vs acquired (pituitary tumor, opioid-induced, functional). MONDO models none of the acquired forms; recipient MONDO:0015770 exists.
  • Parkinson disease — monogenic PD vs idiopathic PD vs secondary parkinsonism (drug-induced, manganism, vascular). dismech models this correctly already, and better than MONDO does, by grouping Manganism with idiopathic PD.
  • bronchiectasis, preeclampsia, temporal lobe epilepsy, keratoconus — each has a small gene-defined arm and a large acquired/idiopathic arm that MONDO does not model.

The generalisable rule: the OMIMPS names the genetic arm; the MONDO class it is mapped to names the whole clinical entity. Every one of the 42 defects is an instance of that single mismatch, and both the dismech question and the Mondo question reduce to it. In dismech terms — the genetic arm is the grouping, the clinical entity is a Disease, and where the genetic arm is a set of risk loci rather than of diseases, there is no grouping at all.

Changes made in this PR

  • scripts/omimps_grouping_audit.py — the reproducible audit.
  • docs/reports/data/omimps-phenotypic-series-heterogeneity-2026-08-01.tsv — full per-class table.
  • kb/groupings/Hedgehog_Pathway_Holoprosencephaly.yaml — worked example of a SHARED_PATHWAY grouping minted over the genetic subset: SHH (ligand), GLI2 (terminal transcriptional activator), CNOT1 (post-transcriptional regulator of SHH repression), with the non-genetic causes of holoprosencephaly — maternal pregestational diabetes, teratogens, trisomy 13 — deliberately outside the boundary. All three members audit SATISFIED against the AND(holoprosencephaly, smoothened signalling) criterion.
  • kb/groupings/Congenital_and_Syndromic_Hypogonadotropic_Hypogonadism.yaml — worked example of the harder case: a shared final common mechanism (deficient GnRH drive) reached by two routes the OMIM series and the MONDO class do not distinguish — developmental (Kallmann/ANOS1, FGFR1: GnRH neurons never arrive or never signal) versus degenerative (Gordon Holmes/RNF216, Boucher-Neuhauser/PNPLA6: normal puberty, then loss) — plus three syndromic members. That split is the thing the grouping records and neither source does.
  • kb/disorders/FGFR1_Hypogonadotropic_Hypogonadism.yaml — a curation defect the grouping audit surfaced: the entry annotated its cardinal phenotype with the parent term HP:0000135 (Hypogonadism) despite hypogonadotropic hypogonadism being its defining feature. Refined to HP:0000044, supported by a GeneReviews sentence already cited elsewhere in the same file.
  • kb/disorders/Celiac_Disease.yaml — the worked example of the recommended shape: the ten GWAS risk genes gain gene_term HGNC bindings and relationship_type: SUSCEPTIBILITY, and the HLA-DQ2/DQ8 haplotype rows gain relationship_type: RISK_FACTOR with a note recording that they are the CELIAC1 (OMIM:212750, HLA-DQA1/HLA-DQB1 at 6p21.32) locus of PS212750. No new claims are asserted — relationship_type is a controlled restatement of the existing free-text association, and the term bindings are validated by just validate-terms.

Follow-ups

  • dismech: two of the four recommended groupings are built here (holoprosencephaly, hypogonadotropic hypogonadism). Remaining: fetal akinesia / arthrogryposis multiplex congenita (Marden-Walker, Wieacker-Wolff, CHRNA1 fetal hypokinesia, X-linked infantile SMA, and Schaaf-Yang, which reaches it through distal arthrogryposis — note dismech already holds an Arthrogryposis Multiplex Congenita umbrella Disease entry, so this one takes the Diabetes_Mellitus shape of Grouping-beside-umbrella-Disease); and monogenic Parkinson disease, best done by extending the existing Parkinsonism_Dopaminergic_Degeneration_Disorders grouping (currently idiopathic PD + Manganism) rather than minting a new one, since a fresh grouping would silently re-lump the acquired member. Separately, enrich the risk-locus blocks of the other SINGLE_DISEASE entries dismech already holds. Two curation gaps the grouping audits surfaced: Schaaf-Yang Syndrome documents hypogonadism without a gonadotropin profile (retained as a member and flagged, not silently admitted), and SHH_Holoprosencephaly_Spectrum has four subtypes with no subtype_term bindings.
  • Mondo: (i) drop autism from the PATTERN list — it is correctly placed under inherited disease susceptibility; (ii) for the 16 susceptibility/locus series, STRIP rather than PATTERN — there is no genetic subset to group; (iii) for immunodeficiency disease, craniosynostosis, hypogonadotropic hypogonadism, hyperinsulinemic hypoglycemia and paraganglioma the recipient class already exists (MONDO:0003778, MONDO:0015338, MONDO:0015770, MONDO:0017182, MONDO:0017366) — relocate rather than mint; (iv) re-rank the work by materialized superclass damage, not descendant count, or single-descendant classes like Waldenström macroglobulinemia (currently an inborn error of metabolism) stay at the bottom of the list.