this threshold in order to exclude those variants with minimal allele ΔF among human populations. Then, performing separate analyses for AAs and EAs and for the Yale–Penn and SAGE datasets, we estimated the association of rs10031423, rs1693457 and rs6846835 with AD symptom counts in accordance with two different models using the R package genome-wide association/interaction analysis and rare variant analysis with family data (GWAF) to fit a generalized estimating equations model to correct for correlations among related individuals [33]. The first model (‘A’) tested the association of the imputed minor allele dosage with the DSM-IV symptom counts for AD considered as phenotype and using DSM-IV cocaine dependence symptom counts, DSM-IV OD symptom counts, DSM-IV ND symptom counts, sex, age and the first three ancestry principal components, as covariates. The second model (‘B’) performed the same analysis with the addition of a further covariate, a variant with ΔF>0.10. Then, we meta-analyzed the results obtained in the Yale–Penn and SAGE datasets for each ancestry group, applying the following equations: βMETA=(βYale-Penn∗WYale-Penn+βSAGE∗WSAGE)/(WYale-Penn+WSAGE) where βMETA, βYale–Penn and βSAGE are the β values in the meta-analysis, Yale–Penn and SAGE datasets, respectively. The meta-analyzed p-values were calculated using METAL software [34]. To estimate the effect of each
