Identification of pyroptosis-associated genes for the prediction of metabolic dysfunction-associated steatohepatitis based on interpretable machine learning models

BACKGROUND: Pyroptosis, a pro-inflammatory form of regulated cell death mediated by gasdermin pore formation and typically triggered by inflammasome activation, has been increasingly recognized as an important contributor to liver inflammation and fibrosis in metabolic dysfunction-associated steatohepatitis (MASH). Despite accumulating evidence linking pyroptosis to MASH pathogenesis, the diagnostic value of pyroptosis-related genes in this disease remains largely undefined. Therefore, the present study aims to identify key pyroptosis-associated molecular signatures with potential utility for the diagnosis of MASH. METHODS: Transcriptomic datasets and corresponding clinical information for MASH patients and healthy individuals were retrieved from the Gene Expression Omnibus (GEO) database. Differential expression analysis using the Limma package, followed by pathway enrichment analyses, was conducted to identify pyroptosis-related genes associated with MASH. Machine learning approaches were applied to systematically screen for core pyroptosis-associated markers and construct predictive models for MASH diagnosis. The robustness of selected gene signatures was further validated in independent datasets and in vivo animal models and vitro cellular models. Prognostic risk assessment was performed using a nomogram informed by key pyroptosis-related genes. Additionally, molecular subtyping of MASH based on pyroptosis gene expression profiles was explored to delineate disease heterogeneity. RESULTS: Through integrative bioinformatics and machine learning, five principal pyro-related genes-LPL, FABP4, STMN2, AKR1B10 and EEF1A2-were identified in MASH. Validation studies in animal model and cell culture systems confirmed the differential expression patterns of these genes. Among evaluated algorithms, Random Forest achieved the highest AUC (0.957) for diagnostic performance. All the five symbols were subsequently included in logistic regression and nomogram models, both demonstrating strong predictive value for MASH diagnosis. Molecular subtyping uncovered substantial variation in pyroptosis gene signatures, immune microenvironment characteristics, and pathway enrichment across MASH subgroups. CONCLUSION: This study highlights the relevance of pyroptosis-related gene signatures in MASH, providing a basis for enhanced diagnostic accuracy and paving the way for individualized therapeutic interventions targeting disease subtypes.