Objective: A growing body of emerging evidence indicates that metabolic processes play a pivotal role in the biological processes underlying acute myocardial infarction (AMI). The aim of the current study was to identify featured metabolism-related genes in patients with AMI using a support vector machine (SVM) and to further explore the value of these genes in the diagnosis of AMI.
Methods: Gene microarray expression data related to AMI were downloaded from the GSE66360 dataset in the Gene Expression Omnibus (GEO) database. This data set consisted of 50 AMI samples and 49 normal controls that were randomly classified into a discovery cohort (21 AMI samples and 22 normal controls) and a validation cohort (28 AMI and 28 normal controls). We applied a machine learning method that combined SVM with recursive feature elimination (RFE) to discriminate AMI patients from normal controls. Based on this, an SVM classifier was constructed. Receiver operating characteristic (ROC) analysis was used to investigate the predictive value for the early diagnosis of AMI in the two cohorts and was then further verified in an independent external cohort.
Results: Three metabolism-related genes were identified based on SVM-RFE (AKR1C3, GLUL, and PDE4B). The SVM classifier based on the three genes allowed for excellent discrimination between AMI and healthy samples in both the discovery cohort (AUC = 0.989) and the validation cohort (AUC = 0.964), and this was further confirmed in the GSE68060 dataset (AUC = 0.839). Additionally, the SVM classifier allowed for perfect discrimination between recurrent AMI events and nonrecurrent events in the GSE68060 cohort (AUC = 0.992). GO and KEGG pathway enrichment analysis of the identified featured genes revealed significant enrichment of specific metabolic pathways.
Conclusion: The identified metabolism-related genes may play important roles in the development of AMI and may represent diagnostic and therapeutic biomarkers of AMI.
Copyright © 2020 Hang Xie et al.