Genotyping was performed using the Illumina 2.5M array, Illumina OmniExpress, Illumina 1M array, or the Affymetrix Smokescreen array; quality control and imputation have been previously described [42]. SNPs with a genotyping rate <98% or that violated Hardy-Weinberg equilibrium (p < 10−6), or with minor allele frequency (MAF) less than 3% were excluded from analyses. Mendelian inconsistencies were removed, after which data were imputed to 1000 genomes (EUR and AFR, Phase 3, b37, October 2014; build hg19) using SHAPEIT [43] and IMPUTE2 [44]. Following imputation, genotype probabilities ≥0.90 were changed to genotypes. Mendelian errors in the imputed SNPs were reviewed and resolved as described previously [42]. SNPs with an imputation information score <0.30 or MAF < 0.03 were excluded from subsequent analysis. Based on 1000genomesv3, principal components (PCs) derived from GWAS data were used to ‘assign’ ancestry in the genotyped sample. We generated polygenic risk scores (PRS) for problematic alcohol use from the largest available GWAS meta-analysis (with COGA data removed) [45]. We constructed PRS using PRS-CSx, which applies a Bayesian regression with a continuous shrinkage parameter to GWAS summary statistics from multiple populations simultaneously [46, 47]. PRS were computed separately for each group of European ancestry and African ancestry participants.
