Target Metabolome Profiling-Based Machine Learning as a Diagnostic Approach for Cardiovascular Diseases in Adults

Abstract

Metabolomics is a promising technology for the application of translational medicine to cardiovascular risk. Here, we applied a liquid chromatography/tandem mass spectrometry approach to explore the associations between plasma concentrations of amino acids, methylarginines, acylcarnitines, and tryptophan catabolism metabolites and cardiometabolic risk factors in patients diagnosed with arterial hypertension (HTA) (n = 61), coronary artery disease (CAD) (n = 48), and non-cardiovascular disease (CVD) individuals (n = 27). In total, almost all significantly different acylcarnitines, amino acids, methylarginines, and intermediates of the kynurenic and indolic tryptophan conversion pathways presented increased (p < 0.05) in concentration levels during the progression of CVD, indicating an association of inflammation, mitochondrial imbalance, and oxidative stress with early stages of CVD. Additionally, the random forest algorithm was found to have the highest prediction power in multiclass and binary classification patients with CAD, HTA, and non-CVD individuals and globally between CVD and non-CVD individuals (accuracy equal to 0.80 and 0.91, respectively). Thus, the present study provided a complex approach for the risk stratification of patients with CAD, patients with HTA, and non-CVD individuals using targeted metabolomics profiling.