Harnessing Large-Scale Multi-Omics Data for Risk Prediction and Deep Phenotyping of Valvular Heart Diseases in the General Population.

The risk profile of valvular heart disease (VHD) and its underlying mechanisms remain poorly understood. This study aimed to develop and validate a multi-omics-based risk prediction model, and to elucidate potential biological mechanisms. Using data from the UK Biobank, Cox proportional hazards and machine learning models (XGBoost and LightGBM) were evaluated for predicting VHD and its subtypes (aortic valve stenosis, AVS; aortic valve regurgitation, AVR; mitral valve regurgitation, MVR). Cox models based on key clinical factors showed the best predictive performance (C-index of 0.75-0.81), which was further enhanced by incorporating proteomic data (all C-index > 0.81) but not by genomic or metabolomic data. Notably, a simplified 10-year model comprising only four top proteins maintained favorable performance (C-index of 0.75-0.82). Cluster analysis identified blood pressure and lipid levels as leading modifiable risk factors for VHD onset. Functional enrichment analysis revealed that VHD is primarily associated with protease inhibition, AVS with fibrotic and matrix metabolic pathways, and MVR with immune-inflammatory activation. Mendelian randomization and Bayesian colocalization analyses suggested causal associations between CNTN5 and CD8A with risks of AVS and MVR, whilst IGFBP7 showed a reverse-direction association with AVS. These findings highlight promising avenues for early diagnostic biomarkers and potential precision-targeted therapies.