Abstract
Genomics, metabolomics, and proteomics offer complementary insights into cardiovascular disease (CVD) risk. Leveraging UK Biobank data, we introduce the CardiOmicScore, a multitask deep learning framework, to learn disease-specific proteomic (ProScore) and metabolomic (MetScore) risk scores for the six most common CVDs by profiling 2920 proteins and 168 metabolites. Experiments demonstrate that ProScore and MetScore are strong sole CVD risk predictors (C-index range: 0.69-0.82 for ProScore and 0.64-0.74 for MetScore), and can significantly enhance risk prediction across CVDs up to 15 years prior to disease onset when combined with clinical data, increasing the C-index by 0.005-0.102. These findings suggest that incorporating multiomics profiling into clinical practice can improve personalized risk assessments at early stages. CardiOmicScore also identifies important CVD-related proteins and metabolites, which represent promising data-driven pathways, calling for further external validation, to develop novel biomarkers and targeted therapies, facilitating precision medicine for primary prevention of CVDs.