samtools
GL genotype ordering does not account for phasing
GL genotype ordering defines genotypes on the a lower triangular matrix of possible alleles, order by GT ordinal.
This is problematic for scenarios where phased GL is calculated as part of the call itself. In such scenario, 0|1 and 1|0 have separate likelihoods.
How do we define GL defined in such scenarios? Presumably we have to define GL of 0/1 as 0|1 + 1|0, but that's a lossy operation.
Similarly, how is GQ defined for a phased variant? Is the genotype the genotype in the GT field or is the the score of the unphased genotype?
We seem to have an implicit assumption that phasing is only ever done after genotyping.
I would assume we simply have to interpret GL as the likelihood that those alleles are the correct ones, and if we want to be sure that the phasing is correct then we should have a separate phasing likelihood score. They are two different things after all, and I don't see the benefit of combining them in a single score.
