Dr. Yin, an Assistant Professor at the University of Florida with postdoctoral training from Harvard Medical School, introduces ADSMoE—a novel mixture-of-experts transformer autoencoder designed to better capture the complex heterogeneity of Alzheimer’s disease (AD). Unlike traditional methods that struggle with incomplete or single-modality-dominated real-world data, ADSMoE integrates neuroimaging, genetic, protein, and clinical data to jointly learn cross-modal signatures. Validated across three large cohorts (ADNI, NACC, and UKBB), the framework successfully identified four distinct, biologically meaningful AD subtypes, offering a scalable solution for patient stratification and subtype-tailored clinical studies.
Presenters