Polygenic risk scores are constructed using SNP effect sizes estimated from genome-wide association studies, which perform marginal regression of the phenotype of interest on each SNP in turn. Explicitly, for continuous traits, we estimate effect sizes b^i (where i = 1,…,M indexes genetic markers) using the model y = b0 + bigi + bPCPC + ε, where gi denotes genotypes at marker i, PC denotes one or more principal components used to adjust for ancestry, and ε denotes environmental noise. For binary traits, we use the analogous logistic model logit[P(y=1) ] = b0 + bigi + bPCPC + ε.
