Can we leverage metafounder gamma matrix in AlphaPeel?

[gregorgorjanc]

We have support for genetic groups / metafounders in AlphaPeel so that we can provide or estimate group specific allele frequencies based on the data below them, which then trickles back down when we peel down and up ...

But, there is an alternative idea, where one sets base population allele freqs to 0.5 and infers relatedness between the members of the base population (=genetic group, metafounder), giving the "gamma" value or matrix in case of several base populations. This comes from the single-step GBLUP literature - to make the pedigree-based relatedness matrix A comparable with the genotype-based relatedness matrix G for the genotyped individuals (so the A22 part of A) - we estimate the gamma values that maximizes the "comparability" of A22 and G (loosely speaking). Would having such a gamma functionality help AlphaPeel too (this would be an alternative to estimating base pop allele frequencies)? The idea is that estimating base pop allele frequencies is challenging due to past selection for which we might not have enough data showing this selection, while when matching A22 and G we tend to have more information.

This is somewhat related to AlphaImpute2 population-based phasing and imputation - there we work with nominally unrelated base population and try to leverage shared allele/haplotype/genotype information among these individuals.