COGA is a family study that includes both genotypic and alcohol-related phenotypic data [22, 23]. Genotyping and imputation were previously described [24]. We selected variants identified as PSI-predictive in the CMC elastic net models. We focused on 8,038 European American (EA) individuals from 1127 independent families, the largest ancestry group in COGA. The phenotypes used in this analysis were DSM-IV alcohol dependence (1 if dependent and 0 if non-dependent) and symptom count (SXCT, the number of DSM-IV criteria met by a participant; range from 0 to 7) [24]. We adjusted for 11 covariates: sex, 3 genotyping array platforms, 4 principal components of population stratification, and 3 birth cohorts [24]. Because the COGA genotyping arrays differed from CMC, the imputation might result in different variants. When imputing the PSI of the skipped exon from the COGA genotype data, we only used the variants that were present in both datasets, which was over 90% of the original CMC variants.
