Transcriptomic and integrative machine learning analysis reveals specific circulating microrna signatures in acute myocardial infarction

Abstract

Circulating microRNAs (miRNAs) are increasingly recognized as sensitive, minimally invasive biomarkers for acute myocardial infarction (AMI), but clinical translation is hindered by variability in analytical approaches and limited use of integrative multivariate models. This study aimed to develop a reliable miRNA panel from small RNA sequencing data of platelet-poor plasma for discriminating AMI from stable angina pectoris (SAP), establish a reproducible workflow from discovery to validation, and evaluate whether selected miRNAs may also serve as prognostic biomarkers for early prediction of LVEF reduction (LVEF<50%) one year post-AMI. 574 patients (286 AMI, 288 SAP) were divided into discovery (n=200) and validation (n=374) cohorts. In the discovery phase, platelet-poor plasma samples underwent small RNA sequencing, followed by bioinformatics analyses in R, including DESeq2, sPLS-DA, and LASSO regression with cross-validation and bootstrap resampling. Candidate panels were evaluated using ROC analysis and validated by qPCR. Multivariable logistic regression was applied to assess the ability of selected miRNAs to predict depressed LVEF (LVEF<50%) one year post-AMI. A five-miRNA panel (miR-133a-3p, miR-208b-3p, miR-378a-3p, miR-143-3p, miR-99a-5p) showed excellent discrimination between AMI and SAP in both cohorts (AUC=0.95; 95%CI:0.93–0.98 and AUC=0.95; 95%CI:0.92–0.97). Additionally, miR-143-3p, with baseline LVEF, independently predicted preserved function (LVEF≥50%) 12 months post-AMI (AUC=0.847, 95%CI:0.782–0.912). The sequencing-derived miRNA signature in platelet-poor plasma demonstrated high diagnostic accuracy and cross-platform reproducibility, while miR-143-3p added prognostic value, supporting its potential role in clinically applicable biomarker panels.