We need to incorporate @ravwojdyla's PC-Relate work.
I'm going to wait on that though until we get @tomwhite's take on dispatching and the overall organization.
Note: The biggest remainder from Rafal's work is that PCs for PC-Relate are supposed to be generated using unrelated individuals. GENESIS does this with a first-order approximation to kinship using KING (which doesn't account for admixture like PC-Relate does) followed by PCA on unrelated people, and then finally projection for all people to create PCs for PC-Relate. Hail ignores these parts, or doesn't provide anything for them, so we'll have to decide what our first-order kinship equivalent is, and if it needs to be robust to population structure. GRM/IBD within self-reported ancestral groups may be enough given that we'll always have that info in human GWAS based on high profile datasets.
We need to incorporate @ravwojdyla's PC-Relate work.
I'm going to wait on that though until we get @tomwhite's take on dispatching and the overall organization.
Note: The biggest remainder from Rafal's work is that PCs for PC-Relate are supposed to be generated using unrelated individuals. GENESIS does this with a first-order approximation to kinship using KING (which doesn't account for admixture like PC-Relate does) followed by PCA on unrelated people, and then finally projection for all people to create PCs for PC-Relate. Hail ignores these parts, or doesn't provide anything for them, so we'll have to decide what our first-order kinship equivalent is, and if it needs to be robust to population structure. GRM/IBD within self-reported ancestral groups may be enough given that we'll always have that info in human GWAS based on high profile datasets.