Sex-mismatch signal from chrX/chrY as an independent chimerism cross-check

[davmlaw]

Idea

In sex-mismatched transplants the sex chromosomes carry a strong, independent chimerism signal that we currently throw away. A colleague raised it: "it seems a shame not to employ that in the fraction of mismatched transplants." Roughly half of transplants are sex-mismatched, and we often know donor/recipient sex up front, so this is more actionable for us than it is for forensic tools.

This is deliberately separate from #46. That issue recovers chrX markers for sex-matched pairs, where the ploidy matches and the markers slot into the existing diploid MLE unchanged. This issue is about the mismatch case, which is a different measurement and should not be folded into the autosomal fraction estimate.

Why keep it separate from the MLE

1. chrY is a different unit. chrY has essentially no heterozygous SNPs; the engraftment signal is normalized read depth (chrY-vs-autosome coverage ratio), not a REF/ALT allele fraction. It does not fit the beta-binomial VAF likelihood. Mixing depth-based and VAF-based terms into one likelihood is fragile to calibrate.
2. chrX in a mismatch breaks the /2 dosage math. The male is hemizygous (1 chrX copy) vs 2 in the female, so the read fraction at a chrX marker is weighted by copy-number x cellularity, not cellularity alone. The chrX "f" is not the same f as the autosomal f. Handling it means per-sample ploidy and a copy-number-weighted mixture term, i.e. mutating the validated diploid core (PLOIDY, ref_dose, the /2).
3. The value is its orthogonality. A sex-chrom estimate fails in completely different ways than the SNP MLE. Reported as an independent estimate alongside the fraction, it becomes a genuine cross-check: agreement raises confidence; disagreement flags a problem (sample swap, contamination, wrong sex metadata). Averaging it into the headline number destroys exactly that.

Reference point

Demixtify (Crysup & Woerner, the basis for our MLE) is explicitly autosomal-only: the README states it "is only for interpretting autosomal biallelic SNPs" and every bundled panel is chr1-chr22. So there is no upstream model to copy for sex chroms; this would be a deliberate, HSCT-specific addition.

Proposed shape

• A separate module, only active when host and donor sex are known/inferred to differ.
• Primary readout: chrY depth ratio (clean binary/semi-quantitative engraftment-direction signal; direction depends on which of host/donor is male).
• Optional secondary: chrX copy-number-weighted estimate.
• Present the result next to the MLE fraction as an independent confirmation + a concordance QC check. Do not blend it into the reported chimerism percentage.

Prerequisites / caveats

• Needs sex inference (shared with chrX marker recovery for sex-matched host/donor pairs #46).
• Panel-agnostic: only works if the input panel actually carries chrX/chrY amplicons; our deployment panel has ~27 chrX, chrY content to be checked. Degrade gracefully when absent.
• Exclude PAR1/PAR2 (diploid in both sexes).

Evidence: [speculative]. Lower priority than the sex-matched recovery in #46, but higher-value per marker because the signal is orthogonal.