Machine learning and artificial intelligence tools were used to investigate the discriminatory potential of blood serum metabolites for thromboembolism and antiphospholipid syndrome (APS). 1H-NMR-based metabonomics data of the serum samples of patients with arterial or venous thromboembolism (VTE) without APS (n = 32), thrombotic primary APS patients (APS, n = 32), and healthy controls (HCs) (n = 32) were investigated. Unique metabolic profiles between VTE and HCs, APS and HCs, and between VTE and triple-positive APS groups were indicative of the significant alterations in the metabolic pathways of glycolysis, the TCA cycle, lipid metabolism, and branched-chain amino acid (BCAA) metabolism, and pointed to the complex pathogenesis mechanisms of APS and VTE. Histidine, 3-hydroxybutyrate, and threonine were shown to be the top three metabolites with the most substantial impact on model predictions, suggesting that these metabolites play a pivotal role in distinguishing among APS, VTE, and HCs. These metabolites might be potential biomarkers to differentiate APS and VTE patients.
Keywords: antiphospholipid antibodies (aPLs); antiphospholipid syndrome (APS); machine learning; metabonomics; thrombosis; venous thromboembolism (VTE).