Alzheimer’s Disease (AD) presents significant challenges in prevention and treatment despite decades of research advancements. Innovative AI/ML approaches enable analysis of real-world data sources, such as electronic health records (EHRs) and longitudinal multimodal data to derive insights without the constraints of predefined selection criteria. Recent developments in integrative heterogeneous graph databases enable the synthesis of knowledge across omics relationships, facilitating the identification of molecular hypotheses linked to complex clinical phenotypes.

National Institute on Aging performed deep phenotyping to characterize AD and sex differences in the EHR compared to a control cohort. They identified sex and AD associated with comorbidities, medication use, and lab results. They employed ML techniques to predict AD onset using clinical information and identify prioritized genes through knowledge network (e.g., APOE, ACTB, IL6) and genetic colocalization analysis (e.g., MS4A6A with osteoporosis). Findings indicate that clinical data can effectively predict the risk of AD onset while highlighting sex-specific relationships before disease manifestation. This work has paved the way for current approaches that leverage unsupervised learning and LLMs to elucidate AD heterogeneity further with the goal of facilitating advances in personalized prediction and interventions for AD.