Multi-task learning identifies shared genetic risk for late-onset epilepsy and alzheimer's disease

Aging populations face increasing incidence of neurological disorders, including Alzheimer's disease (AD) and late-onset epilepsy (LOE), which demonstrate a bidirectional relationship where AD is a risk factor for LOE and LOE is a risk factor for AD. While the APOE gene is a known shared risk factor, comprehensive genetic studies for LOE remain limited. This study employed a multi-task learning framework using Elastic Net modeling to systematically identify shared genetic risk factors between AD and LOE. We analyzed electronic health records from UCLA Health System (N = 416,212; genetic subset N = 16,500) and validated findings in the All of Us dataset (N = 52,493). Longitudinal analyses confirmed strong bidirectional associations between AD and LOE. The multi-task learning approach identified eight shared-risk single nucleotide polymorphisms mapping to key genes including the APOE-TOMM40-APOC1 cluster, BIN1, CLU, PVRL2, and TRAPPC6A. These shared-risk genes were enriched in pathways related to lipid metabolism, amyloid catabolic processes, and tau protein binding. A shared genetic risk score effectively stratified patients into distinct AD-LOE risk groups. This study represents an initial systematic identification of potential shared genetic factors between AD and LOE using multi-task learning. While our findings suggest possible shared genetic contributions, particularly in the APOE region, and highlight tau-mediated mechanisms as potential therapeutic targets, further validation is needed to establish the extent of genetic overlap between these conditions.