Trio analysis — validated
vcfclick's inheritance models aren't checked against hand-built fixtures alone. They're validated against the Genome in a Bottle Ashkenazi trio — HG002 (son), HG003 (father), HG004 (mother) — using GIAB's high-confidence BED as an independent ground truth, and cross-checked against two established tools.
The de-novo defensibility test
De novo means the child carries a variant neither parent has — which needs the parents to be provably hom-reference, not just absent from a sparse genotype table. In a 5 Mb region of chr20, HG002 carries 50 variants absent from both parents. The naive "neither parent carries" rule calls all 50 de novo. GIAB's BED says otherwise:
50
naive de-novo candidates
HG002-only variants in 5 Mb
4
confirmed de novo
both parents in-BED (confident 0/0)
46
false positives avoided
a parent was a no-call, not 0/0
vcfclick's --keep-reference de novo
recovers exactly those 4 — a ~12× over-call eliminated by requiring stored,
provable parent hom-reference.
The 50/4/46 split is from a one-time scan of the 5 Mb region against GIAB's BED; the automated test exercises a representative 7-site subset (the 4 confirmed + 3 no-call).
Three independent tools agree
On the same real GIAB genotypes, vcfclick, bcftools +mendelian2, and slivar (the standard rare-disease trio tool) flag the same de-novo sites — and all three exclude the
no-call-parent sites for the same reason (a ./. parent is not a
confident 0/0).
| de-novo site | vcfclick | bcftools +mendelian2 | slivar |
|---|---|---|---|
| chr20:2202238 | ✓ | ✓ | ✓ |
| chr20:5739093 | ✓ | ✓ | ✓ |
| chr20:5864738 | ✓ | ✓ | ✓ |
| chr20:5893991 | ✓ | ✓ | ✓ |
| 3 no-call-parent sites | excluded | excluded | excluded |
Recessive on the CFTR region likewise matches: vcfclick and slivar both flag exactly
chr7:117,488,888 and chr7:117,507,446. The bcftools +mendelian2 agreement is asserted in CI; the slivar
comparison is a documented one-time run (slivar ships a Linux binary only).
Every model, on real genotypes
In a real GIAB CFTR slice the benchmark genotypes reproduce each model: two recessive sites (proband hom-alt, both parents het), three dominant single-origin hets, and CFTR as a genuine compound-het candidate gene — a paternal-origin het plus maternal-origin hets in trans.
Population-frequency filtering
With real gnomAD v4.1 frequencies loaded, --gnomad-max-af 0.01 cuts those
CFTR candidates sharply — the recessive sites are common in gnomAD (popmax ≈ 0.8), while one
dominant het is genuinely rare (popmax 0.009). Defensible rarity from true population data,
not the cohort's own frequencies.