Multi-Omics Integration Improves Polygenic Risk Prediction for Lipid Traits: A Multi-Ancestry Study in UK Biobank

Background: Polygenic risk scores (PRS) have proven valuable for disease risk prediction, but their predictive utility often remains limited because human traits result from complex interactions between environmental and genetic factors. Blood lipid levels are heritable and clinically important risk factors for cardiovascular disease, yet it remains unclear whether multi-omics integration can enhance lipid trait prediction beyond PRS alone. Methods: We first constructed single-omics scores, where gene expression, plasma protein, and plasma/serum metabolite levels were genetically predicted and weighted by effect sizes estimated via LASSO regression. Subsequently, we implemented two integration strategies to develop composite multi-omics risk scores (MoRS): step-MoRS, which integrates single-omics scores using stepwise regression, and Lasso-MoRS, which directly models all predicted features across omics layers using LASSO regression. Both approaches were evaluated across European, South Asian, and African ancestries within the UK Biobank. Results: MoRS-based methods consistently demonstrated superior predictive accuracy compared to PRS alone for four lipid traits across diverse populations. Notably, Lasso-MoRS prioritized key biomarkers with predictive utility complementary to genomic data. Conclusions: These findings confirm that integrating multi-omics biomarkers with genomic data significantly enhances lipid trait prediction across diverse ancestries, offering biological insights into the molecular regulation of lipid metabolism.