in-sample, optimization was done within a single PRS to ensure limited overfitting of the combined model.10 We used 5-fold cross-validation to assess model performance in which 80% of the cohort was used to estimate the mixing coefficients (α1 and α2) and the combined PRS partial-R2 was calculated in the remaining 20% of the data. Reported partial-R2 was averaged across folds.10 For our binary phenotypes with unbalanced affected and unaffected individuals, we used stratified 5-fold cross-validation.
