Unsupervised Deep Embedded Clustering Reveals High-Risk Subgroups for Alzheimer’s Disease

Alzheimer’s disease (AD) is a complex and heterogeneous condition influenced by diverse and evolving risk factors, complicating prevention, detection, and treatment. Profiling trajectories of high-risk individuals offers a promising strategy to advance precision medicine (PM). The Wisconsin Registry for Alzheimer’s Prevention (WRAP) provides over 9 years of longitudinal data, including diagnoses, medications, labs, and imaging for at-risk individuals. This study investigated whether distinct WRAP subgroups exist to stratify AD risk. Using Deep Embedded Clustering (DEC), an advanced technique combining dimensionality reduction and iterative clustering, we analyzed 398 participants. DEC identified six clinically distinct subgroups, outperforming traditional methods (n = 2) in granularity and relevance. Cluster 1 (Healthy) had the highest cognitive scores (PACC-4: 1.1) and optimal metabolic markers. Cluster 2 (Early-Risk) showed intermediate metabolic markers and white matter hyperintensity lesions (WMHs). Cluster 3 (Advanced-AD) had the lowest PACC-4 scores (−0.2) and elevated GFAP levels (145.4 pg/mL), indicating severe neurodegeneration. Cluster 4 (Genetic-Risk) was enriched with APOE ε4 carriers (47.5%) and high amyloid-beta ratios. Cluster 5 (Vascular Contributions) demonstrated high LDL cholesterol (119.1 mg/dL) and WMHs, while Cluster 6 (Inflammatory Burden) exhibited systemic inflammation and insulin dysregulation. Clusters were validated using independent metrics, including cognitive performance, plasma biomarkers, MRI imaging, and medication patterns. Robustness was confirmed through Jaccard stability, entropy, and Shapley value analysis, identifying key drivers of cluster membership. DEC effectively stratifies heterogeneous AD populations, advancing PM through tailored interventions targeting metabolic, genetic, vascular, and inflammatory pathways.