Stratifying Alzheimer’s disease by patient-specific genetic signatures reveals cognition-linked and cross-disease heterogeneity

Abstract

Alzheimer’s disease (AD) presents profound clinical and genetic heterogeneity that obscures its biological underpinnings and impedes therapeutic development. While genome-wide studies have identified population-level risk loci, the architecture of patient-specific genetic variation and its link to clinical outcomes remains poorly defined. Here, we introduce a heteroscedastic personalized regression (Het-PR) framework to move beyond cohort-averaged associations and construct individualized single-nucleotide polymorphism (SNP)-effect profiles for each subject. Applying this method to the Alzheimer’s Disease Neuroimaging Initiative (ADNI) cohort, we identify an internally stable, exploratory genetic-profile stratification of AD patients into two subgroups that exhibit divergent performance across five cognitive domains. This finding suggests an association between model-derived individualized genetic profiles and cognitive impairment severity. Cohort-level analysis confirms that frequently selected variants map to biologically relevant, brain-expressed genes. We further find that genetic variants previously associated with multiple neuropsychiatric and cognitive traits distinguish the AD subgroups under a label-permutation enrichment analysis, with epilepsy-associated variants showing the strongest proportional signal among the tested trait categories. These results suggest that AD heterogeneity may reflect shared genetic architecture across broader brain-related traits, including neuronal excitability-related loci such as sodium voltage-gated channel alpha subunit 1 (SCN1A). Our results provide an exploratory, genetically informed framework for studying AD heterogeneity, showing that model-derived individualized SNP-score profiles can reveal latent structure associated with cognitive and cross-disease genetic patterns.