The companion to apoe_epsilon, probing the half of RM28 that APOE left open.
APOE showed a two-SNP haplotype needs no predicate. This asks the harder question: compound
heterozygosity — where the same two alleles mean opposite things depending on which chromosome they
sit on — was the case that most justified a predicate language. Does it need one?
It does not. And building it surfaced the thing that is actually missing, which turned out to be a compiler check rather than a schema feature.
HFE haemochromatosis is recessive. Two alleles matter in practice:
| allele | variant | GRCh38 |
|---|---|---|
| C282Y | rs1800562 G>A | 6:26092913 |
| H63D | rs1799945 C>G | 6:26090951 |
C282Y homozygosity accounts for most clinically diagnosed cases. C282Y/H63D compound heterozygosity — one variant on each chromosome — is a recognised at-risk genotype: no wild-type HFE protein is made from either copy, and a minority develop raised iron indices. Carrying both variants on the same chromosome is a different situation entirely: the other chromosome still carries an intact HFE, so the expectation is that of a simple C282Y carrier.
Same two variants. Same two heterozygous calls. Opposite finding.
A diplotype is already a statement about two homologs. haplotypes.csv says which alleles ride
together on one chromosome (same-strand conjunction, as APOE showed), and diplotypes.csv pairs two
of them — which is exactly what "in trans" means. So cis and trans are two rows:
| haplotype | rs1800562 | rs1799945 |
|---|---|---|
wt |
G | C |
C282Y |
A | C |
H63D |
G | G |
C282Y-H63D |
A | G |
C282Y/H63D→ trans: compound heterozygote, at-risk.C282Y-H63D/wt→ cis: both variants on one chromosome, one intact copy, a carrier.
Both compile today with bricks that shipped in 0.4. The predicate grammar RM28 sketched would have restated this less legibly, and the cis/trans relation it was going to introduce is what a diplotype pair is.
Note wt is written out with its reference alleles at both positions, the same choice ε3 makes in
the APOE example — unlike a star-allele *1, which is defined by carrying no variants at all.
Nothing in the module says that a consumer cannot tell those two rows apart. Discard phase and both present the identical genotype — rs1800562 G/A and rs1799945 C/G — with opposite conclusions. Almost all consumer genotype data is unphased. Silently reporting the first would manufacture an at-risk finding; silently reporting the second would suppress one.
That is real, and it is derivable: the compiler already holds both tables, and the computation is
pure and offline. So it became a check rather than a column —
compiler._cross_validate_phase_ambiguity, a warning that never blocks:
warning: HFE: diplotype rows C282Y-H63D/wt, C282Y/H63D are indistinguishable without phase —
they present the same unphased genotype but state different conclusions. A consumer with unphased
calls must withhold rather than pick one; a phased consumer resolves it.
A requires_phase column would have made an author restate what the data already determines, and it
would go stale the moment a haplotype is edited. Compiling this module emits exactly one such warning
and nothing else; the other six diplotype rows are unambiguous.
The check is closed-world: it compares the rows a module states, never the rows it omits. APOE is the illustration — ε2/ε4 and ε1/ε3 are the textbook unphased collision, and because that module carries no ε1, nothing fires. That is correct (the module makes no claim about ε1), and the neighbouring "star allele used but not defined" warning is what covers an allele a caller might emit that the module never describes.
And the two gaps apoe_epsilon named are untouched by this:
- Pairing across subjects — a conclusion combining an HFE diplotype with, say, a HAMP or TFR2 variant has no carrier; every table keys on one subject.
- Economy. Four haplotypes give ten diplotypes and that is comfortable. A gene with 300 pathogenic variants where "any two in trans" is the finding gives ~45,000 pairs — expressible, unwritable, and no longer saying why once enumerated.
Neither is an operator problem, which is the point ROADMAP RM28 keeps making: rows are a disjunction and columns are a conjunction, so the tables already span any finite boolean function over an enumerable set of genotypes. What is missing is economy and open-world negation, and no predicate token fixes the second.
clin_sig follows ClinVar's own calls for the two variants, and the conclusions are deliberately
hedged: HFE penetrance is incomplete and strongly modified by sex, age, alcohol and blood loss, so
every row reads as a susceptibility statement rather than a diagnosis. ACMG SF v3.2 lists HFE, and
scopes it to "c.845G>A; p.C282Y homozygotes only" — narrower than the gene, and narrower than
this module, which is why no row here claims secondary-findings reportability.
just-dna-enricher enrich reference_examples/hfe_compound_het
just-dna-compiler compile reference_examples/hfe_compound_het out/hfe_compound_het