Given a pair of polygenic risk scores computed as above using two distinct training populations, we define the multi-ethnic PRS with mixing weights α1 and α2 as the linear combination of the two PRS with these weights: e.g., for EUR and LAT, we define PRSEUR+LAT = α1PRSEUR + α2PRSLAT. We employ two different approaches to avoid overfitting. In our primary analyses, we estimate mixing weights α1 and α2 using validation data and compute adjusted R2 to account for the additional degree of freedom. In our secondary analyses, we estimate mixing weights α1 and α2 using cross-validation (see Assessment of methods below).
