Phenocopies

Scope

Issue #10395 asks for a phenocopy presentation on rare genetic disease entries: where an exposure can produce what looks like a Mendelian disease in someone carrying none of its variants, show the exposure and show what tells the two apart.

The clinical observation is not in dispute. What is open is whether dismech needs a new construct to record it, and if so which.

Two senses of the word, and they must not be merged. "Phenocopy" already appears throughout kb/, almost entirely in the experimental sense: a chemical or knockdown reproducing a mutant phenotype in a model system. This project concerns the clinical/diagnostic sense. The two carry different evidence rules, and a slot added without naming the distinction will fill with model-system results.

Source artefacts

The entries and modules declared in this page's frontmatter are the dismech content the compilation's rows touch, whether or not the text below names them individually. They are a reading list, not a worklist, and nothing has been curated against them.

The compilation's own caveats govern. It states that tiers are a judgement about mechanistic tightness and not an evidence rating, that two rows are arguably gene–environment interaction rather than phenocopy, that valproate/POLG is deliberately both a phenocopy and a trigger and would need two edge types, and that the set is a literature compilation rather than curated content. No row has been verified for this project.

A section heading is not a tier count. The five sections group by theme and the tier runs across them, so the two do not line up. The first section, "Identical molecular target", holds 14 rows of which 11 are tier 1; its other three, aminoglycosides, warfarin and the benzocaine/dapsone/nitrites row, are tier 2 and do not inhibit the gene product the section blurb describes. Read the tier column rather than the heading.

Decided 2026-09-24: every row is verified against primary literature before it is curated. The compilation is a starting list, not a source, so no entry is written from a row that has not been checked.

What the knowledge base already expresses

dismech already carries phenocopy pairs, and in one case already models the relation.

The construct is DifferentialDiagnosis, reached through differential_diagnoses, carrying name, disease_term, description, phenotypes, distinguishing_features, evidence and notes.

So the question is narrower than the issue implies. It is not whether dismech can express this, but that:

• the relation is untyped, so the phenocopy claim lives in free text and is not queryable;
• the relation is one-directional, so a pair is discoverable from one side only;
• the mechanistic tightness distinction, between sharing a molecular target and merely converging on a presentation, has nowhere structured to go.

Implementation options

From comment 5497102936, cheapest first. All three are undecided.

That comment's argument for schema content over prose is that where an exposure hits the same gene product the mutation hits, the phenocopy is not really a differential diagnosis but a second etiologic route into an existing pathograph node. Assess that claim before choosing, since it is what separates option B from option A.

A. Fields on DifferentialDiagnosis. An etiology_class, an optional ECTO/CHEBI exposure_term, and a relation grade. Existing differential blocks stay valid, no new section, no new foreign-key surface, and the view becomes a render-time filter on the existing differentials card. That last point rests on the disorder page being one long page of anchored section cards rather than a tab widget, which is unverified.

B. A top-level phenocopies: section with attaches_to, distinguishing_features, discriminating_tests, reversibility and evidence. Richer, but a new section and a new foreign-key surface. Better deferred until A has produced a corpus to design against.

C. A Grouping at kb/groupings/Toxic_Phenocopies_of_Genetic_Disease.yaml. No schema change, following the Digenic_and_Oligogenic_Disorders pattern.

Open questions

1. Does a typed qualifier on DifferentialDiagnosis suffice? A negative answer removes most of the remaining work. Settle this first.
2. What counts as verified? The decision above settles that rows are verified, not what the check covers. The pair and its shared lesion are the minimum; whether it also covers the tier assignment and the distinguishing test changes the cost across 48 rows.
3. One tier scheme or two? The compilation proposes T1/T2/T3; the issue comment proposes mechanistic, pathway and phenotypic grades. They were written independently and appear to mean nearly the same thing. One should win, and neither may be rendered anywhere it could read as a confidence grade.
4. Where does the relation attach? Disease to disease, exposure to disease, or exposure to a named pathophysiology node.
5. Directionality. Recorded on the genetic entry, the toxic entry, or both. dismech's convention for cross-entry claims is duplication rather than inheritance, which argues for both, each written from its own point of view.
6. Is discriminating_tests the highest-value field? No existing construct has it. The compilation already supplies it for all 48 rows.
7. How is reversibility handled? 20 of the 48 pairs halt or regress on withdrawal, which is diagnostically decisive and has no home in the schema.
8. What about pairs that are both phenocopy and interaction? Valproate/POLG is the worked case.

Relation to the AOP work

Issue #10395 notes the overlap with the AOP work, and it is real but narrower than first framed. An agent asserted that a toxic phenocopy is an AOP whose adverse outcome is a genetic disease's phenotype set. That does not hold: a phenocopy could correspond to an AO at the end of an AOP, but shouldn't be conflated with an entire AOP.

An AOP defines a causal series of biological events leading to an adverse outcome (AO). An AO is often an adverse phenotype, but only by virtue of biology rather than by definition. The definition is tied to the purpose AOPs were designed to serve, which is regulatory: an AO is defined by its relevance as a regulatory endpoint informing chemical safety decision-making. That said, the framework has since been adopted for organizing evidence on causal biological sequences outside regulatory science.

What the two bodies of work do share is the question of where a disease-level correspondence attaches, since it names a whole entry rather than any single node. That is already open as question 2 in AOP_EMOD_ALIGNMENT and is tracked there rather than here.

Not yet included

• Any proposed slot name or enum.
• Any rendering design.
• Verification of the 48 rows, or of the pairs listed above.