To choose the optimal predictive set of SNP’s on the target data set, we conducted the standard LD clumping followed by P value thresholding procedure. The LD clumping was done on windows of 250 kb with a squared correlation of allele counts r2 = 0.1. This means that within a given 250 kb window with r2 = 0.1, the SNP with the smallest p value was chosen as a representative SNP. A PTSD-PRS was constructed and its performance (Nagelkerke R2, a measure of coefficient of determination for binary traits) was evaluated over a grid of ten equally spaced p value thresholds from 0.1 to 1 inclusive, and the nominal significance threshold of 0.05 (Fig. S1). Each time only SNP’s with lower P value than the threshold were included in the PRS summation and the p value threshold with the best performance (PT = 0.2) was chosen for the final PRS calculation. For schizophrenia–PRS, a threshold of 0.05 had been shown to be the most predictive in the original publication27. Thus, this threshold was used to avoid multiple testing burden. The
