For unrelated case-control samples, we used the logistic regression analysis implemented in PLINK to test associations between genotypes and phenotypes, with ancestry proportions (Zuo et al., 2012), sex, age and alcohol drinking behavior (i.e., at least 12 alcoholic drinks in the past 12 months) as covariates; for family samples, we used FBAT to test associations. The cleaned marker numbers, the minor allele frequencies (MAFs) of the most significant risk SNPs and the minimal p values, are shown in Table S26. The significance levels (α) were corrected for the number of effective markers (calculated by a Bonferroni-type program SNPSpD that takes LD structure into account (Li and Ji, 2005)) and the number of cohorts (i.e., 21). The number of risk SNPs that were nominally (p<0.05) or significantly (p<α) associated with phenotypes are shown in Table S27. Finally, an overall π0 value for associations with multiple comparisons was estimated from p values within each cohort using the R package QVALUE (Storey and Tibshirani, 2003). A π0 value is the overall proportion of true null hypotheses, reflecting the overall behavior of the estimated
