We used the pruning and thresholding method because it can select SNVs that have the largest contributions for the purpose of developing a specific array for AUD PGS calculation to eliminate genotyping platform effects. First, PLINK, version 2,43,44 was used to select sets of independent concordant SNVs by varying P value thresholds (>.99, .50, .20, .10, .05, .01, .005, .001, .0005, .0001, .00005,.00001, .000005, .000001, .0000005, .0000001, and .00000005), linkage disequilibrium (LD) r2 values (0.1, 0.2, 0.3, 0.4, and 0.5), and physical distance to calculate r2 (250 kb and 500 kb), resulting in 170 sets of SNVs. LD was determined using 1000 Genomes Project samples of European ancestry. For individual i, PGS was calculated as PGSi = ΣM(j=1) βj × dosageij, where M is the number of SNVs and βj is the z score for SNV j estimated from meta-analysis of GWAS of MVP, UKBB, and FinnGen; dosageij is the number of effective alleles in the AOU dataset and imputation dosage in the IB dataset for individual i for SNV j.
