Integrated metabolomics and machine learning for coronary artery disease severity stratification: identification of...

Coronary artery disease (CAD) remains a leading cause of global mortality and morbidity. Current diagnostic methods, such as coronary angiography, are invasive and resource intensive. Metabolomics offers a powerful approach to identify disease-related metabolic signatures. This study aims to characterize plasma metabolic alterations in CAD patients and develop a biomarker panel for severity stratification. A total of 617 participants, including 297 healthy controls and 320 CAD patients, are classified by coronary angiography into mild (< 70% stenosis) and severe (≥ 70% stenosis) groups and further stratified by clinical presentation into asymptomatic, angina, and myocardial infarction subgroups. Untargeted metabolomics is performed on plasma samples. Differential metabolites are identified via statistical analysis, and machine learning models are developed to predict CAD severity. A total of 253 significantly altered metabolites are identified between CAD patients and controls, which are primarily involved in lipid metabolism, caffeine metabolism, amino acid metabolism, and energy pathways. Key metabolites such as caffeine show stepwise decreases with increasing CAD severity, while oxylipins exhibit progressive increases. Weighted gene co-expression network analysis (WGCNA) identifies 30 metabolic modules, with modules significantly correlated with CAD severity. Machine learning models integrating clinical and metabolomic data achieve high diagnostic accuracy, with area under the receiver operating characteristic curve (AUC) values > 0.95 for CAD stratification. Therefore, this study comprehensively characterizes plasma metabolic alterations in CAD and identifies potential biomarker panels for clinical stratification. The integration of metabolomics and machine learning shows promise for the early detection and personalized management of CAD.