Work

Machine Learning identification in patients with aortic valve stenosis of a myocardial miRNA signature predictive of reverse cardiac remodelling after valve replacement

Year: 2026

Type: preprint

Abstract: ABSTRACT Background Aortic stenosis (AS) is a valvular heart disease which imposes chronic pressure overload on the left ventricle (LV). Although LV remodelling may regress following aortic valve replacement, the molecular determinants of reverse remodelling remain poorly understood. This study develops a machine learning-based model using myocardial microRNA (miRNA) expression profiles to predict LV mass normalization one year after valve replacement. Methods Consensus selection combining Random Forest and Bayesian Network analyses was applied to miRNA sequencing data from LV biopsies of 12 matched patient pairs with AS. Selected miRNAs were qPCR quantified in an extended cohort of 89 AS patients and incorporated into predictive models. Sex-stratified analyses were performed. Model robustness was assessed using bootstrap resampling and cross-validation. Candidate miRNAs and target genes were evaluated in LV from a murine model of pressure overload induced by transverse aortic constriction (TAC). Results The combined Random Forest–Bayesian Network approach identified a three-miRNA signature (miR-517a-3p, miR-122b-5p, and miR-4683) associated with LV mass normalization after valve replacement. Validation in an extensive cohort demonstrated good overall discrimination and achieved its highest performance in women (AUC >0.80). Functional analyses identified drebrin (DBN1) as a putative target of miR-517a-3p and supported its potential involvement in pressure overload-induced cardiac remodelling. Conclusion Random Forest and Bayesian Network integrated analyses enabled the identification of a myocardial miRNA signature with sex-dependent predictive value for reverse cardiac remodelling after aortic valve replacement. Functional validation supports the relevance of selected miRNAs and their downstream targets, highlighting their potential as biomarkers and candidate regulators of myocardial recovery. What Is New? In surgical LV myocardial biopsies from AS patients we detected, using machine learning strategies, miRnoma changes prognostic of the LV postoperative reverse remodeling completion. Sex stratified predictive models reveal biologically meaningful differences that could refine prognostic modelling and future therapeutic targeting in reverse cardiac remodelling. Low LV miR 517a 3p and high DBN1 expressions associated persistent one-year postoperative hypertrophy in patients, a pattern consistent with post transcriptional regulation that appeared also in pressure overloaded and de-loaded mice. What Are the Clinical Implications? Preoperative information on the post-surgical fate of LV remodelling in patients with severe AS may be important for the therapeutic decision-making process, particularly in asymptomatic patients. A favourable prognostic miRNA signature might be an argument to defer valve implantation (open or TAVR) in an asymptomatic patient with with severe AS and a procedural high-risk profile. Conversely, a worsening miRNA signature might prompt valve replacement in a well fit asymptomatic AS patient. The bedside availability of this information will be key for its incorporation into decision algorithms.

Source: medRxiv

DOI: 10.64898/2026.09.22.26363736