Trio/Duo inheritance mode for a mosaic parent (dominant, low-VAF parent)

[davmlaw]

🤖 Written by Claude

A medical scientist asked whether a dominant node can find variants where a parent is mosaic - the child is constitutional HET (~50% VAF) and the parent carries the variant in a fraction of cells. In the data the parent is either (a) called 0/0 by the germline caller with a handful of alt reads in AD (below the ~15-20% VAF calling threshold - the common case), or (b) called HET with a low VAF (5-30%).

What works today

Trio node in Denovo mode, then add the parents' Allele depth / Allele frequency sample columns to the grid and sort by them - candidates with parental alt reads float to the top. Covers case (a) with no code: CohortGenotype.samples_allele_depth / samples_allele_frequency are stored per sample per variant, and in a joint-called multi-sample VCF the parent's values at the proband's site are stored even when its GT is 0/0.

The gap

Case (b): Denovo excludes a HET-called parent, and Dominant with an unaffected parent does too. The Trio node's AD/DP/GQ/PL thresholds and the Allele Frequency slider apply one range across all samples (ANY/ALL), so "proband AF ≥ 0.3 AND parent AF 0.02-0.35" can't be expressed.

Proposal - TrioInheritance.MOSAIC_PARENT "Dominant (mosaic parent)"

• proband: HAS_VARIANT
• one parent (either, OR'd): zygosity unconstrained (HOM_REF / HET / MISSING) and (samples_allele_frequency in [min, max] or samples_allele_depth ≥ N)
• other parent: NO_VARIANT and AF below min
• two node fields, mosaic_max_af (default ~0.35) and mosaic_min_alt_reads (default 2), shown as a row in the zygosity table
• same mode on the Duo node (Duo analysis: proband + one parent (Duo model, wizard and DuoNode) #1829) with its single parent - the "other parent clean" clause drops

AD is the robust signal (AF is only stored when the VCF has an AF FORMAT field); samples_allele_depth is the alt depth so AD ≥ N is a direct array-index filter like the AF one in cohort_mixin.get_allele_frequency_q_list.

Warnings the node should carry

• Needs a joint-called (multi-sample) VCF - per-sample VCFs merged into a cohort have no parent record at the proband's site, so no AD.
• Depth: a 5% mosaic at 30x is ~1.5 reads. Meaningful at ~100x+ / targeted panels; on standard WGS results are mostly sequencing noise and index hopping. Sorting by parent AD is the honest presentation.
• Blood mosaicism ≠ gonadal - absent signal in blood doesn't exclude germline mosaicism. A recurrence-risk hint, not a rule-out.