We used 10-fold cross-validation to account for overfitting. That is, for each of 10 iterations, polygenic scores were first derived from LD-pruned estimates of single SNP effects in a training sub-sample comprising 90% of the overall sample. Training sample estimates were then used to predict the phenotypes of individuals in a separate testing sub-sample comprising the remaining 10% of the overall sample. Then, at each p-value threshold, polygenic score-based predictions were compared to the actual observed phenotypic values of individuals in the testing sub-sample using the coefficient of determination, averaged across all 10 iterations of the cross-validation procedure.
