For comparison purposes in analyses of real phenotypes, we also evaluated a meta-analysis PRS (e.g. EUR-LAT-meta) using a sample size weighted average of estimated effect sizes in each population (Zeggini et al., 2008); for dichotomous phenotypes we weighted by effective sample size Neff=4/(1/Ncase+1/Ncontrol). We performed LD-pruning and P-value thresholding using P-values obtained from the meta-analysis, using the LD reference panel from the population that achieved the highest prediction accuracy.
