Improving Genetic Risk Prediction Across Diverse Population By Disentangling Ancestry Representations

DisPred is a deep-learning-based framework that can integrate data from diverse populations to improve the generalizability of genetic risk prediction. It combines a disentangling approach to separate the effect of ancestry from the phenotype-specific representation, and an ensemble modeling approach to combine the predictions from disentangled latent representation and original data. DisPred captures non-linear genotype-phenotype relationships without restricting specific ancestral composition and does not require self-reported ancestry information for predicting future individuals. Evaluation of DisPred performance to predict AD risk prediction in a multi-ethnic cohort composed of AD cases and controls showed that DisPred performs better than existing models in minority populations, particularly for admixed individuals.