Identification and Functional Characterization of COPD Molecular Subtypes Based on Oxeiptosis-Related Genes via WGCNA and Machine Learning

Background: This study establishes the first molecular stratification framework for chronic obstructive pulmonary disease (COPD) based on oxeiptosis biology, exploring how this reactive oxygen species (ROS)-induced cell death pathway shapes disease heterogeneity. The precise role of oxeiptosis in COPD pathogenesis remains poorly understood. Methods: Transcriptomic profiles from two independent cohorts, GSE47460 (220 COPD cases and 108 controls) and GSE76925 (111 COPD cases and 40 controls), were systematically evaluated using a fully in silico computational pipeline. Oxeiptosis-related differentially expressed genes (ORDEGs) were identified through correlation and differential expression analyses, prioritized using machine learning, and then applied for unsupervised clustering. Subtypes were externally validated and functionally characterized through pathway enrichment and network analysis. Results: The support vector machine (SVM) model prioritized four of the seven ORDEGs for further analysis. Two reproducible subtypes were identified: C1, characterized by diminished ORDEGs activity and significantly worse pulmonary function (FEV 1 % predicted: C1 vs. C2, 44% vs 59%, P < 0.05), and C2, marked by heightened activity and less severe dysfunction. Weighted gene coexpression network analysis (WGCNA) revealed a black module associated with subtype classification, enriched in cytokine signaling and extracellular matrix remodeling pathways. A predictive model was constructed to investigate clinical applicability. Conclusion: This computational discovery framework introduces an oxeiptosis-specific molecular taxonomy of COPD. The results underscore oxidative stress and the interaction between immune regulation and matrix remodeling as pivotal elements linked to disease heterogeneity, presenting potential pathways for precise diagnosis and treatment. Keywords: COPD, oxeiptosis, oxidative stress, molecular subtypes, machine learning