Investigation of potential biomarkers in prediction of acute myocardial infarction via explainable artificial intelligence

Abstract

Remodeling of the left ventricle (LV) after myocardial infarction (MI) is a process of infarct enlargement. Despite the relevance of the inflammatory response and healing process in LV remodeling after MI, the mechanisms that begin and govern these processes remain unknown. Based on the important information highlighted in different studies, the current research aims to investigate potential biomarkers for left ventricular remodeling after acute MI based on the interpretation of the explainable artificial intelligence (XAI). The project research from which the public dataset was obtained was designed in an experimental type. A cohort study involving 66 patients with coronary heart disease and 34 healthy community controls provided the platelet samples for the current research, which used available omics data on those samples. For discovering significant mechanistic connections between metabolites and glycans, the metabolomics and glycomics datasets were analyzed using biostatistics/metabolomics and explainable artificial intelligence techniques. Metabolomics data of 100 patients (AMI=66; Control=34) including 75 males and 25 females were evaluated in this study. As a result of experimental omics analyses, 102 metabolite levels of the patients were obtained. When FC values were examined, creatinine and dl-pipecolic acid levels were 0.50 and 0.55-fold down-regulated and glutamine, myoinositol, and cytosine levels were 1.34, 1.33, and 1.53-fold up-regulated in the AMI group compared to the control group. Findings of metabolomics data and XAI analyses revealed that five lipid metabolites may be used as potential predictors of AMI.