Identification and clinical evaluation of diagnostic biomarkers for ischemic cardiomyopathy based on machine learning and transcriptomics.

Background: Ischemic cardiomyopathy (ICM) is a leading cause of heart failure, yet precise molecular tools for potential diagnosis remain limited. This study aims to identify and clinically evaluate novel diagnostic biomarkers for ICM by integrating comprehensive transcriptomic analysis, machine learning algorithms, and real-world serological assessment. Methods: Gene expression profiles from the GEO database were systematically analyzed using weighted gene co-expression network analysis (WGCNA) and 12 distinct machine learning algorithms to screen for optimal diagnostic targets. Crucially, to bridge the gap between computational prediction and clinical application, the identified core diagnostic genes were evaluated at the protein level using an independent clinical cohort. Peripheral serum samples from 90 individuals (45 ICM patients and 45 controls) were analyzed via enzyme-linked immunosorbent assay (ELISA). Results: We identified 501 differentially expressed genes, with the MEcyan WGCNA module showing the strongest correlation with ICM. Among the 12 evaluated machine learning models, AdaBoost showed the highest internal predictive performance (AUC = 0.976). However, this estimate is exploratory and potentially optimistic due to feature pre-selection prior to cross-validation. From this candidate pool, SEPP1, CILP, and FRZB were selected Conclusion: By integrating computational screening with preliminary clinical serological analysis, this study identifies SEPP1, CILP, and FRZB as potential serological biomarkers for ICM. This computational model and the accompanying preliminary serological evidence provide a theoretical basis for future clinical exploration and diagnostic biomarker development.