Chimerism report

At markers where every donor is homozygous for the same allele and the host carries the other allele, that donor-absent allele appears only at the sequencing-error background in a pure-donor sample. A one-sided test (a likelihood ratio, with a pooled-Poisson cross-check) asks whether the observed donor-absent reads exceed that background, that is, whether any host signal is present.

This answers "is the host detectable?" directly, separately from the size of the donor fraction estimated above, and is most informative near the detection limit.

Goodness-of-fit compares the observed allele depths to those predicted by the fitted fraction; a low p-value means the single-fraction model fits poorly (an unexpected marker, an extra contributor, or a genotyping error).

Contamination measures third-party DNA at markers where the host and every donor share the same homozygote: minor-allele reads there cannot come from either contributor, so their excess over the sequencing-error floor estimates contamination (for example index hopping or cross-contamination).

Sample-swap counts those same shared-homozygote sites where the minor allele is individually significant; many such sites indicate a whole extra genome (a swap or wrong-patient VCF).

Shared-het balance checks the admixture allele fraction at markers heterozygous in the host and every donor, where it should sit near 50% whatever the mixing fraction. Many sites skewed away from balance point to contamination, copy-number or allelic imbalance, or a sample mix-up.

Relatedness is a kinship coefficient between the reference samples, compared against the declared relationship to catch mislabelled or unexpectedly related inputs.

The robust refit re-fits after dropping markers whose residuals are large outliers; dropping a large fraction is itself flagged, since it points at host copy-number change or a genotyping problem.