Dilated cardiomyopathy signature metabolic marker screening: Machine learning and multi-omics analysis

Objects: Our aim was to identify changes in the metabolome in dilated cardiomyopathy (DCM) as well as to construct a metabolic diagnostic model for DCM. Methods: We utilized non-targeted metabolomics with a cross-sectional cohort of age- and sex-matched DCM patients and controls. Metabolomics data were analyzed using orthogonal partial least squares-discriminant analysis (OPLS-DA) and pathway analysis. It was validated in combination with transcriptome sequencing data from public databases. Machine learning models were used for the diagnosis of DCM. Results: Using multiple analytical techniques, 130 metabolite alterations were identified in DCM compared to healthy controls. Perturbations in glycerophospholipid metabolism (GPL) were identified and validated as a characteristic metabolic pathway in DCM. Through the least absolute shrinkage and selection operator (LASSO), we identified the 7 most important GPL metabolites, including LysoPA (16:0/0:0), LysoPA (18:1(9Z)/0:0), PC (20:3(8Z,11Z,14Z)/20:1(11Z)), PC (20:0/14:0), LysoPC (16:0), PS(15:0/18:0), and PE(16:0/20:4 (5Z,8Z,11Z,14Z)). The machine learning models based on the seven metabolites all had good accuracy in distinguishing DCM [All area under the curve (AUC) > 0.900], and the artificial neural network (ANN) model performed the most consistently (AUC: 0.919 ± 0.075). Conclusions: This study demonstrates that GPL metabolism may play a contributing role in the pathophysiological mechanisms of DCM. The 7-GPL metabolite model may help for early diagnosis of DCM.