indicate higher accuracy scores for EA and AA males, when the ancestry was defined by genetics than by self-report for models with only PRS features. Several discriminative features were identified for each of the models revealing novel predisposition sex and ancestry-specific AUD remission biomarkers. EEG-FC in all groups was found to distinguish between the continued AUD and the remitted AUD group, revealing DMN and fusiform lower and insula higher functional connectivity in the continued AUD group. Several discriminative PRS were shared (neuroticism PRS and years of education PRS in EA groups), while others PRS were group-specific, such as PRS associated with aggression were important for EA males, and depression PRS were important for EA females. Being married, employed, and taking medication predicted the maintenance of the AUD state. Overall, our findings suggest that wide range of multidimensional features with high internal homogeneity groups will formulate better predictive models.
