DisSNPNet: Predicting disease-associated single-nucleotide polymorphisms using linkage disequilibrium, disease similarity, and 1000 Genomes Project datasets with evidence-based validation

Abstract Identifying disease-associated single-nucleotide polymorphisms (SNPs) is fundamental to understanding complex disease genetics, yet genome-wide association studies (GWAS) remain costly and data-intensive. Network-based approaches provide a complementary strategy by exploiting linkage disequilibrium (LD) structure- and disease-relatedness to prioritize candidate variants. We present DisSNPNet, a heterogeneous network-based framework that integrates chromosome-specific SNP LD networks derived from 1000 Genomes Project Phase 1 and Phase 3 data, a MeSH-based disease similarity network, and known disease–SNP associations from CAUSALdb. Random walk with restart was applied to rank SNPs for each disease. Predictive performance was evaluated using disease-wise 3-fold cross-validation with AUROC and AUPR. Biological plausibility was assessed by querying top-ranked SNPs in GWAS resources and by disease-specific KEGG pathway enrichment. A chromosome-matched random baseline was constructed to contextualize external GWAS evidence. DisSNPNet consistently outperformed SNP-only LD networks, with heterogeneous networks yielding higher AUROC and AUPR across chromosomes. Strong LD networks (r2 ≥ 0.8) improved precision, particularly in imbalanced settings. Top-ranked SNPs showed significantly greater GWAS evidence than random expectation across all chromosomes, indicating nonrandom enrichment. Disease-specific pathway enrichment revealed biologically coherent mechanisms across immune, metabolic, cardiovascular, and structural diseases. DisSNPNet provides a robust and interpretable framework for prioritizing disease-associated SNPs. While not a substitute for GWAS, it offers a scalable, evidence-supported approach for SNP prioritization and hypothesis generation, complementing experimental and population-based studies.