We performed out-of-sample prediction using individual-level genotype and phenotype data from the PGC29-MDD cohorts5. We obtained permissions from 20 cohorts with sample sizes greater than 500, among which 17 recorded endorsement of DSM-5 criteria A for MDD (Supplementary Note and Supplementary Table 21). We obtained PRSs from GWAS for each definition of depression in the UK Biobank, using LD-clumped (LD r2 < 0.1) independent SNPs with P values for association below eight thresholds (P < 10−4, 0.001, 0.01, 0.05, 0.1, 0.2, 0.5 and 1), and predicted MDD status in the 20 PGC cohorts using the Ricopili pipeline70-82. We obtained Nagelkerke’s r2 between the PRSs and MDD status, the AUC of the prediction and the variance of MDD status explained by the PRSs for each cohort. We also obtained the same measures for MDD status pulling data from all cohorts, controlling for cohort differences by including cohort as a covariate.
