In linear regression, we’ve added the option use_genotypes=True
to use genotype dosage in place of the gt
field (hard call). Dosage is defined as the expected genotype under the additive HomRef=0, Het=1, HomVar=2
model, as derived from genotype probabilities or phred-scaled likelihoods with a uniform prior.
For example, if
P(HomRef) = 0.1, P(Het) = 0.7, P(HomVar) = 0.2
then the dosage is
0 * 0.1 + 1 * 0.7 + 2 * 0.2 = 1.1
See the documentation for details. On deck we have performance improvements and extending the dosage option to logistic and linear mixed regression.