Skip to content

Mitochondrial Disease Genetics and Pathograph Integration Review (2026-08-27)

Review of how dismech assigns genes to mitochondrial disease entries, whether that strategy is complete for mitochondrially encoded genes, and whether the genetics and the core GO biology actually reach the pathographs.

Motivating question: HGNC and OMIM were both suspected of having incomplete classification for mitochondrial genes. HGNC's coverage turns out to be complete; its classification is not usable, and the real gaps are elsewhere.

Scope

102 entries qualify as mitochondrial disease: every entry with a curated causal gene that is mtDNA-encoded or belongs to the nuclear OXPHOS / mtDNA-maintenance / mitochondrial translation / CoQ / mitochondrial dynamics gene set, plus every entry carrying parents: mitochondrial disease. Together they hold 579 pathophysiology nodes and 276 curated genes.

15 of the 102 are mtDNA-encoded (MT-*) disorders; the remaining 87 are nuclear-encoded.

The gene-assignment strategy, stated

Genes are bound as HGNC CURIEs on GeneDescriptor.gene_term, validated by linkml-term-validator against the GeneTerm dynamic enum. GeneTerm carries no reachable_from constraint — unlike CellTypeTerm (rooted at CL:0000000) or the NCIT treatment enums, it validates only that the CURIE exists and that term.label matches the canonical HGNC symbol exactly. CURIEs use the repository's lowercase hgnc: form.

OMIM is not part of gene assignment. Its 1,165 occurrences in kb/ are cross-references (external_assertions, mappings) and prose. So the suspicion about OMIM's mtDNA gene-phenotype map does not affect how genes are assigned — but it does mean OMIM cannot be used to audit completeness, since OMIM lumps most mtDNA phenotypes under allelic variants of a single MIM number rather than distinct phenotype entries.

Finding 1 — HGNC covers all 37 mtDNA genes

Checked directly against the OAK sqlite:obo:hgnc build: all 37 mitochondrially encoded genes resolve with correct labels and SO type assignments.

Class Count Example SO type
Protein-coding 13 hgnc:7414 MT-ATP6 SO:0001217 protein_coding_gene
rRNA 2 hgnc:7470 MT-RNR1 SO:0001637 rRNA_gene
tRNA 22 hgnc:7490 MT-TL1 SO:0001272 tRNA_gene

None missing. HGNC additionally carries 16 non-gene mtDNA control-region features (MT-HSP1, MT-LSP, MT-OHR, MT-TER, MT-CSB1..3, MT-TAS, MT-7SDNA, MT-ATT), which is more than the KB currently needs but is available for D-loop variants.

There is no identifier-level gap. The 14 MT genes in cache/hgnc/terms.csv are simply the ones curated so far, not the limit of what validates.

Finding 2 — HGNC's classification of mtDNA genes is unusable here

The gene groups exist but do not form a coherent axis:

  • hgnc.genegroup:1974 covers only the 13 protein-coding genes.
  • tRNA genes sit in 843, rRNA genes in 1378. No superclass unites all 37.
  • In the OAK sqlite build every hgnc.genegroup:* node is a label-less stub carrying only rdf:type — no rdfs:label, no group hierarchy.

So a reachable_from: hgnc.genegroup:1974 dynamic enum would yield an unlabeled, incomplete set. HGNC cannot answer "is this gene mtDNA-encoded" for dismech.

Nor can SO: MT-ND1 and NDUFS4 are both SO:0001217. Genome of origin is not derivable from any authority dismech currently binds to. Today the only signal is the MT- symbol prefix, which is a naming convention, not an assertion.

Finding 3 — heteroplasmy is unmodeled

For an mtDNA disease, heteroplasmy fraction and the tissue-specific threshold are the genetic parameters that determine penetrance and severity. dismech has no slot for either:

  • ZygosityEnum is HETEROZYGOUS | SIMPLE_HETEROZYGOUS | COMPOUND_HETEROZYGOUS | HOMOZYGOUS | HEMIZYGOUS — no homoplasmic/heteroplasmic values, and zygosity is orthogonal to heteroplasmy in any case.
  • The word "heteroplasmy" appears in 20 files, always as free text in description or notes — 41 times in MT-ATP6_MT-ATP8-Related_Infantile_Hypertrophic_Cardiomyopathy alone.

This is a real representational gap, not a curation lapse: curators wrote the biology into prose because there was nowhere else to put it. See "Recommendation" below; it is a structural schema decision and is proposed, not enacted, here.

Finding 4 — 42% of curated genes never reached the pathograph

graph.py links a genetic[] entry to a mechanism only by matching gene keys against pathophysiology[].genes / .gene (_gene_lookup_keys, lines 176–191, 371–390). A pathophysiology node with no genes: is therefore invisible to the genetic block, and the gene renders as a disconnected node.

Before this review:

Measure Value
Pathophysiology nodes carrying no genes: 492 / 579 (85%)
Curated genes not reachable from any node 116 / 276 (42%)
Entries with ≥1 orphaned gene 54
Entries whose entire genetic block was disconnected 40

Among the 40 were entries named for their gene — COX10-Related_COX_Deficiency, SURF1-Related_Leigh_Syndrome, TACO1-Related_COX_Deficiency — whose proximal node ("COX10 Loss and Defective Heme A Biosynthesis") named the gene in prose while linking nothing.

Finding 5 — GO coverage is solid on OXPHOS, thin elsewhere

273 distinct GO terms are used across the 102 entries; none is obsolete. Coverage concentrates on the respiratory chain and falls away outside it:

GO term Entries
GO:0006119 oxidative phosphorylation 46
GO:0033617 complex IV assembly 22
GO:0032543 mitochondrial translation 13
GO:0007005 mitochondrion organization 11
GO:0006264 mitochondrial DNA replication 9
GO:0008053 mitochondrial fusion 3
GO:0000266 mitochondrial fission 2
GO:0034551 complex III assembly 1
GO:0000423 mitophagy 0
GO:0070585 protein localization to mitochondrion 0
GO:0008637 apoptotic mitochondrial changes 0
GO:0001836 release of cytochrome c from mitochondria 0
GO:0006851 mitochondrial calcium ion transmembrane transport 0
GO:0051881 regulation of mitochondrial membrane potential 0

Zero mitophagy annotation is the most surprising: PRKN-Related_Juvenile_Parkinson_Disease and Parkinson_Disease_Mitochondrial are both in scope. Complex III assembly at 1 entry against complex IV at 22 is a lopsidedness worth a targeted pass.

Also: 152 / 579 nodes (26%) carry no bound biological process, molecular function, or cellular component at all. Most are tissue- or organism-scale "energy failure" nodes (Kearns-Sayre_Syndrome has seven), which could carry GO:0006119 plus an UBERON site rather than being pure free text.

One correction to a plausible-looking gap: GO:0000002 "mitochondrial genome maintenance" is obsolete with no replacement — GO deliberately split it because it conflated transport, lipid metabolism, DNA metabolism, fission, and fusion. The KB's existing use of GO:0006264 (mtDNA replication) and GO:0032042 (mtDNA metabolic process) for the depletion syndromes is correct. GO:0006626 is likewise obsolete, replaced by GO:0070585.

Finding 6 — missing inheritance on mtDNA-caused entries

Five entries with a curated mtDNA cause carried no inheritance: block at all, so nothing recorded maternal transmission. Leigh_Syndrome was the most consequential: six curated genes spanning three inheritance modes (mitochondrial MT-ATP6, autosomal recessive NDUFS4/NDUFV1/SURF1/LRPPRC, X-linked PDHA1) and no inheritance block to distinguish them.

What this review changed

37 KB entries changed, all validated (schema, terms, snippets, duplicate keys, entity refs).

Gene → pathograph links (35 gene links across 34 entries). Attached the curated causal gene to the proximal molecular node that was already named for it. Two passes: the node name contains the HGNC symbol, then the node names the gene product instead (Twinkle → TWNK, DNA Polymerase Gamma → POLG, mt-tRNA(Glu) → MT-TE, COX4-1 → COX4I1). This adds 37 genetic → mechanism edges (16 → 53 across the changed entries). No new biological claims — each gene was already curated with evidence in the same file, and each node was already named for its lesion.

Pathophysiology can carry a gene in three places — a singular gene:, a multivalued genes:, and genetic_context.gene — and _gene_lookup_keys reads only the first two. A node using gene: was therefore already linked, so an initial pass that indexed only genes: proposed 12 edits to nodes already correctly bound; those were withdrawn before merge and are not in this change. A node using genetic_context.gene is the opposite case: it looks bound in the YAML but is invisible to the graph builder, so adding genes: there is a real fix rather than a duplicate (Mitochondrial_Complex_I_Deficiency_Nuclear_Type_31, TIMMDC1, is the worked example — it now states the gene twice, and the second statement is the one the pathograph sees).

Three slots for one fact, two of them read by the graph builder and nothing in the schema, docs, or checks saying which to use, is the underlying defect. It is worth a separate pass: either collapse the slots, or teach _gene_lookup_keys to read genetic_context.gene so the YAML and the graph agree.

Measure Before After
Genes not reachable from any node 116 (42%) 78 (28%)
Entries with ≥1 orphaned gene 54 23
Entries fully disconnected 40 6

Inheritance blocks (4 entries). Leigh_Syndrome (three modes: HP:0001427, HP:0000007, HP:0001417), NARP_syndrome, Adult-Onset_Ataxia_and_Polyneuropathy, Reversible_Infantile_Cytochrome_c_Oxidase_Deficiency (all HP:0001427). Every block carries its own PMID and verified snippet.

The Reversible Infantile COX Deficiency block records the biologically unusual part explicitly: m.14674T>C/T>G are homoplasmic, so there is no heteroplasmy threshold and the reversible course reflects developmental compensation rather than a shifting mutant load.

Recommendation — a structural slot for heteroplasmy and genome of origin

Proposed, not enacted; it is a schema decision for the register rather than a curation fix. Shape:

genetic:
- name: MT-TL1 m.3243A>G
  gene_term: {preferred_term: MT-TL1, term: {id: hgnc:7490, label: MT-TL1}}
  genome: MITOCHONDRIAL          # GenomeEnum: NUCLEAR | MITOCHONDRIAL
  heteroplasmy:
    state: HETEROPLASMIC         # HOMOPLASMIC | HETEROPLASMIC | BOTH
    threshold_percent: 60
    threshold_tissue: skeletal muscle
    evidence: [...]              # standard EvidenceItem

genome is the cheaper half and is mechanically derivable for existing entries (the MT- prefix over a closed 37-gene set), which makes it a safe backfill. heteroplasmy is the half that carries biology no other slot can hold, and it needs per-entry curation from the 20 files that currently state it in prose.

Follow-ups not taken

  • Kearns-Sayre_Syndrome and Pearson_Syndrome have no genetic: block. Both are single large-scale mtDNA deletion syndromes with no single causal gene, so the omission is defensible — but the deletion span is enumerable, and now that all 37 mtDNA genes validate, the genes removed by the common 4,977 bp deletion could be curated explicitly. Needs a reference stating the span; the cached GeneReviews (PMID:20301382) is abstract-only.
  • relationship_type is unset on causal genes in several entries (Leigh_Syndrome all six, MELAS_Syndrome both, NARP_syndrome).
  • 6 entries remain fully disconnected — Charcot-Marie-Tooth_Disease, Complex_Hereditary_Spastic_Paraplegia, Pheochromocytoma_Paraganglioma, Multiple_System_Atrophy, Chronic_Intestinal_Pseudoobstruction, STAT2_Deficiency. All are multi-gene entries where assigning each gene to the right node needs curation judgment, not a mechanical rule.
  • MTO1_Deficiency lists MT-TF as a MODIFIER. MTO1 is a nuclear tRNA-modifying enzyme acting on mt-tRNAs; MT-TF is its substrate, not a modifier locus of MTO1 disease. Worth re-checking against the cited evidence.
  • GO gaps above — mitophagy, mitochondrial protein import, calcium transport, cytochrome c release, complex III assembly.