The GAIN EAs and AAs were analyzed separately. Genotyped SNPs passing quality control were used as input for 1000 Genomes imputation, following the same procedure outlined in the Supplemental Information. A subset of genotyped SNPs in linkage equilibrium (r2<0.5) were used to compute eigenvectors, from which we selected the number needed to explain >90% of the phenotype variance: 4 for GAIN AAs and 6 for GAIN EAs (Table S8). Logistic regression models were run to replicate SNP associations with illicit drug abuse. Additional data were available on alcohol abuse and nicotine dependence in the GAIN cohort. To focus on SNP/indel associations with illicit drug abuse, the replication analysis included sex, selected eigenvectors, alcohol abuse, and nicotine dependence as covariates (see Supplemental Information).
