adolescence, late adolescence and adults) and EA late adolescence age group. Gender analyses revealed trend of higher model accuracy in the female group over the male group in both the EA and the AA for all three features categories (EEG, SNPs, and the combined EEG+SNP model). We further found gender differences in model accuracy with parental history of AUD added to the model. Interestingly, both EA and AA samples showed history of maternal AUD as a discriminative feature, increasing the accuracy of the combined EEG+SNP based model. History of paternal AUD increased the model accuracy over the combined EEG+SNP based model only in the AA females. In both samples, the younger age group achieved higher accuracy score than the two older age groups. Several discriminative EEG and SNP features were identified for each of the models revealing novel gender and ancestry specific AUD predisposition biomarkers. Overall, our findings suggest that higher model accuracy is anchored in a wide range of multidimensional features generated from specific homogenous samples (e.g., gender, age, ancestry). Importantly, identifying group-related specific features will generate formulation of better prediction models.
