Biomarker Identification and Risk Assessment of Cardiovascular Disease Based on Untargeted Metabolomics and Machine Learning

Cardiovascular disease (CVD) is the leading cause of mortality, disability, and healthcare costs, with a significant impact on the elderly and contributing to premature deaths across various age groups, including those below age 70. Despite decades of transformative discoveries and clinical efforts, the challenges of diagnosis, prevention, and treatment of CVD persist on a massive scale. This study aimed to unravel potential CVD-associated biomarkers and establish a machine learning model for the risk assessment of CVD. Untargeted metabolic assay with ultra-high performance liquid chromatography-tandem mass spectrometry and routine clinical biochemistry test were undertaken on the fasting venous blood specimens from 57 subjects. Four relevant clinical traits and 164 CVD-associated metabolites were identified, especially those related to glycerophospholipid metabolism and biosynthesis of unsaturated fatty acids. The machine learning model achieved from an integrated biomarker panel of palmitic amide, oleic acid, 138-pos (the 138th detected metabolomic feature in positive ion mode), phosphatidylcholine, linoleic acid, age, direct bilirubin, and inorganic phosphate, was able to improve the accuracy of CVD risk assessment up to a high satisfactory value of 0.91. The findings indicate that disorders in the metabolic processes of biological membranes and energy are significantly associated with increased risk of vascular damage in CVD patients. With machine learning methods, the pivotal metabolites and clinical biomarkers offer a promising potential for the efficient risk assessment and diagnosis of CVD.