Single-cell machine learning uncovers genetically anchored, cell-type specific programs of Alzheimer’s disease

Abstract Aging and genetic risk shape the molecular programs that confer cellular vulnerability in Alzheimer’s disease (AD), but whether these programs differ between clinical symptoms and neuropathological burden remains unclear. Using single-nucleus RNA sequencing (snRNA-seq) from the dorsolateral prefrontal cortex of 48 donors either with AD or controls (∼70,000 nuclei), we integrated transcriptomic, genetic, and demographic data to identify shared and cell-type–specific molecular predictors of AD. Across six major brain cell types and 41 fine-grained subclusters, we uncovered robust pan-cell-type predictors—including RASGEF1B , LINGO1 , and ARL17B —and distinct cell-specific programs such as CRYAB in oligodendrocytes and IFI44L in microglia. Many of these genes have regulatory links to genome-wide significant AD loci and brain eQTLs, linking genetic susceptibility to transcriptional state. Pseudotime analyses revealed progressive activation of amyloid and neuroinflammatory pathways along disease trajectories, while comparative modeling of clinical versus neuropathological outcomes highlighted divergent molecular programs between symptom manifestation and amyloid plaque burden. Validation in an independent cohort (21 donors, ∼172,000 nuclei) confirmed the reproducibility of predictive features across cell types. By jointly modeling genetic, demographic, and transcriptomic axes, our study nominates high-confidence, genetically anchored molecular drivers of AD and prioritizes them for mechanistic investigation and therapeutic targeting in age-related neurodegenerative disease.