Z. Cheng, S. Judaprawira, J. Perez, A. Goldstein
Sex differences in Alzheimer's disease (AD) genetic effects and regulatory contexts remain incompletely resolved. We analyzed 449,335 European-ancestry participants using sex-stratified and genotype-by-sex models and developed STAGE-AD, a Transformer integrating molecular QTLs, epigenomic and single-cell annotations. Sex-stratified analyses identified eight female and three male genome-wide-significant signals provisionally classified as novel. Primary test-set area under the precision-recall curve was 0.895 (95% confidence interval, 0.877-0.912), decreasing to 0.751 under locus hold-out and 0.681 under APOE-region hold-out. Statistical-genetic integration in independent HUNT and MVP cohorts supported 236 genes at 106 loci. Prespecified lifetime-risk assumptions yielded liability-scale SNP heritabilities of 11.42% in females and 9.23% in males. Ablations indicated that female-biased predictions depended on glial annotations and male-biased predictions on endothelial and oligodendrocyte annotations. Eleven candidate variant-gene-tissue-cell-pathology chains were evaluated using multi-omics and neuropathological data from 579 NPAD brain donors. These findings underscore the importance of considering sex-specific genetic architectures in the study of health conditions, including Alzheimer's disease, paving the way for more targeted treatment strategies.