Abstract / Description
BACKGROUND: Coronary artery disease (CAD) remains the leading cause of global mortality. Multi-omics integration offers unprecedented opportunities for precision risk stratification beyond conventional approaches. However, comprehensive integration of genomic, proteomic, and metabolomic data for CAD prediction remains underexplored. METHODS: We integrated genomic (n = 442,574), proteomic (n = 54,219 with 1472 proteins), and metabolomic (n = 121,249 with 249 metabolites) data from UK Biobank. Multi-omics integration employed similarity network fusion combined with elastic net regularization in the 23,776 participants with complete multi-omics data, with a late integration (stacking) strategy applied for prediction modeling across the full cohort. Internal validation encompassed temporal split (2006-2008 vs. 2009-2010), geographical split (England vs. Scotland/Wales), 10-fold cross-validation, and bootstrap resampling (1000 iterations). RESULTS: Among 442,574 participants (34,241 CAD cases, 7.74%), we identified 241 independent genomic loci and integrated 1472 proteins with 249 metabolites. Network analysis revealed four major biological modules: vascular dysfunction (37.8% variance), lipid metabolism (25.4%), inflammatory response (19.2%), and cardiac remodeling (13.6%). The integrated multi-omics model achieved superior discrimination (C-statistic: 0.798, 95% CI: 0.793-0.803) compared to clinical factors alone (0.741) and genomics-only approaches (0.704, P CONCLUSIONS: Multi-omics integration substantially enhances CAD risk prediction with robust internal validation, enabling targeted precision medicine strategies through successful intermediate-risk reclassification. These findings establish a framework for clinical translation of multi-omics risk assessment.