In order to account for linkage disequilibrium (LD) in the area (Fig. 2), which could confound association results, Bayesian multilocus models were fitted to these data, separately for the three alcohol measures. Figures in Supplementary Material, Table S3 represent the probability that a model with one or more SNPs as explanatory variables is selected out of all the possible models in the model space (statistics based on 10 000 iterations with the first 5000 excluded—‘burn-in’). Results show a strong effect of rs1229984 on alcohol drinking before pregnancy and binge drinking during pregnancy, independent on the other SNPs in the model. However, for one of the outcomes, drinking in the first trimester of pregnancy, the model with rs1229984 performed worse than the null model (posterior probabilities: 0.08 versus 0.88) (Supplementary Material, Table S3). The association between rs2866151 and pre-pregnancy drinking disappears when taking into account LD patterns in this multi-locus model. These results are in line with crude associations from classical univariable analyses (Table 2), confirming that the only SNP associated with alcohol use in this study is the established rs1229984,
