Unveiling unified patterns in Alzheimer's disease subtypes: An SCCA clustering approach integrating PET imaging and genomics data.

Alzheimer's disease (AD) is the most common cause of dementia and a significant public health challenge. AD is characterized by the formation of tau and beta-amyloid (Aβ) protein aggregates in the brain, which can be imaged in vivo using positron emission tomography (PET). Integrating genetic and neuroimaging data using imaging genetics tools offers the potential to better understand disease mechanisms and risk factors in this heterogeneous disorder. Here, we present a framework based on Sparse Canonical Correlation Analysis (SCCA) integrated with clustering to identify AD subtypes from PET and genomic data. The SCCA clustering method was applied to tau PET scans (N = 541), Aβ PET scans (N = 907), and corresponding genomics data from the Alzheimer's Disease Neuroimaging Initiative database. Test-retest studies were used to compare two different SCCA implementations, and longitudinal data were used to assess the stability of the subtyping approach. We identified four tau subtypes and two Aβ subtypes with distinct spatial deposition patterns, consistent with prior imaging studies. Genetic profiles associated with each subtype showed enrichment of specific biological pathways. Our findings suggest that SCCA clustering can help reveal biologically meaningful subtypes of AD. A clearer understanding of AD subtypes could ultimately improve AD diagnosis, prognosis, and treatment strategies.