Clustering Alzheimer’s Disease Subtypes via Similarity Learning and Graph Diffusion

Alzheimer’s disease (AD), a complex neurodegenerative disorder affecting millions globally, presents diagnostic and treatment challenges due to its heterogeneous nature. This study targets the identification of homogeneous AD subtypes with unique clinical and pathological characteristics to overcome these issues. Using an innovative approach with unsupervised clustering, the multi-kernel similarity learning framework SIMLR, and graph diffusion, we analyzed MRI-derived cortical thickness measurements in 829 patients with AD and mild cognitive impairment (MCI). The unique clustering methodology we used, untested in AD subtyping before, outperformed traditional clustering methods, notably mitigating noise interference in subtype detection. Five distinctive subtypes emerged, differing significantly in biomarkers, cognitive status, and other clinical features. A subsequent genetic association study validated these subtypes, uncovering potential genetic connections.