Identifying key genetic variants in Alzheimer’s disease progression using Graph Convolutional Networks (GCN) and biological impact analysis

Abstract Alzheimer’s disease (AD) involves complex genetic interactions that remain challenging to model computationally. We present a novel deep learning framework integrating Single Nucleotide Polymorphism (SNP) data with Graph Convolutional Networks (GCNs) to predict gene-disease relationships in AD. Our dual-pathway architecture combines: (1) linear SNP feature processing for individual genetic variants and (2) non-linear GCN analysis of functional gene networks, fused through an optimized integration module. Using rigorously curated data from the GWAS Catalog and AD-specific functional networks (FGN), the model achieved exceptional performance (accuracy: 98.04 ± 0.32%, AUROC: 0.996). Ablation studies demonstrated statistically significant contributions from both GCN (Δaccuracy − 7.92%, p < 0.001) and SNP pathways (Δaccuracy − 5.74%, p < 0.001), validating their complementary roles in AD prediction. The framework’s biological interpretability revealed known AD risk genes (APOE, PSEN1) while identifying novel network-level associations. This study advances precision medicine in neurodegeneration by providing: (i) a validated tool for early genetic risk assessment, and (ii) mechanistic insights into AD pathogenesis through network medicine paradigms. The model’s modular design permits adaptation to other complex diseases, with immediate applications in clinical trial stratification and therapeutic target discovery.