Within-study prediction accuracy was measured by five-fold cross-validation in a random sampling of 1,000 of the highly heritable genes (i.e. significant non zero cis-heritability) for each study. Cross-study prediction accuracy was measured by merging the YFS/NTR genotyped individuals and predicting from all individuals in one cohort into all individuals in the other. In all instances, the R2 between predicted and true expression across all predicted folds was used to evaluate accuracy (see Supplementary Note).
