We assessed the accuracy of polygenic risk scores in validation samples (independent from samples used to estimate effect sizes). We used adjusted R2 as the accuracy metric for continuous traits and liability-scale adjusted R2 (ref. (Lee, Goddard, Wray, & Visscher, 2012)) for binary traits. Adjusted R2 is defined as R2^−(1−R2^)pn−p−1, where p ∊ {1,2,3} is the number of PRS or ANC components in the mixture, n is the number of validation samples, and R2^ is the raw (unadjusted) R2. The adjusted R2 metric roughly corrects for increased model complexity in multi-component PRS, so in our primary analyses, we report accuracy as adjusted R2 using best-fit mixing weights α^k estimated using the validation data.
