Different SVM prediction models with overlapping features and different subsets of subjects divided according to ancestry, age and gender were used for the predictive models. Table 1 summarizes the results of the significant predictive model scores across ancestry, gender and age (see Supplementary Tables 4–6 for full results), revealing higher scores for the AA than for the EA sample (p < 0.001), for females over males in both EA (borderline trend) and AA (p(EA male vs. female)=0.06, p(AA male vs. female)=0.03) and for the younger age group over the others in both samples (EA; F2 = 76.29, p < 0.001, AA; F2 = 8.27, p = 0.001) (Table 1). Table 1 and Figure 2 highlight that the combined model of EEG+SNP was more accurate than the model based on only EEG features or only SNP features for both the AA and EA samples (EA; p(EEG vs. EEG+SNP)< 0.001, p(SNP vs. EEG+SNP <.001) (AA; p(EEG vs. EEG+SNP)< 0.001, p(SNP vs. EEG+SNP <0.001). Results were confirmed in a follow up analysis in the AA and EA age groups (AA: p(early adolescence ,EEG
