Abstract INTRODUCTION Alzheimer's disease (AD) shows marked molecular heterogeneity. Defining biological subtypes may refine diagnosis and treatment. METHODS We analyzed cerebrospinal fluid (CSF) proteomics and longitudinal data from 550 participants in the Alzheimer's Disease Neuroimaging Initiative with up to 16.5 years of follow‐up. We profiled 6361 proteins, applied machine learning to identify biological subtypes, and validated them in three independent cohorts. RESULTS Three AD subtypes were identified. Subtype 1, enriched in RNA metabolism pathways, showed the mildest atrophy and slowest cognitive decline. Subtype 2, characterized by axonogenesis‐related pathways, exhibited the greatest CSF tau elevations, moderate atrophy, and intermediate decline. Subtype 3, associated with catabolic processes, showed the most severe atrophy and fastest progression. These patterns were consistently replicated across validation cohorts. DISCUSSION These findings demonstrate robust, biologically distinct AD subtypes linked to divergent molecular pathways, clinical features, and progression rates. Such refined stratification supports precision diagnostics and targeted therapeutic strategies.
Hou et al. (2026) studied this question.