Machine Learning Identifies SCO2 as a Metabolic Reprogramming Gene Promoting Lipid Accumulation in MAFLD

Metabolic dysfunction-associated fatty liver disease (MAFLD) is a common chronic liver disease characterized by metabolic dysregulation, inflammation, and fibrosis. Metabolic reprogramming is increasingly recognized as a key molecular driver of MAFLD progression; however, systematic identification of the genes involved remains limited. This study aimed to identify key metabolism reprogramming genes (KMRGs) associated with MAFLD using multi-cohort bioinformatics analyses and machine learning, and to validate their expression patterns and potential roles experimentally. Six MAFLD-related liver transcriptomic datasets were downloaded from the GEO database. GSE135251 and GSE126848 were merged to form the training cohort, whereas the remaining four datasets served as external validation cohorts. Differentially expressed genes (DEGs) were identified and intersected with metabolism reprogramming genes (MRGs) to obtain KMRGs. Functional enrichment analysis, disease association analysis, GeneMANIA network analysis, external validation, immune infiltration analysis, inflammation- and lipid metabolism/fibrosis-related analyses, and consensus clustering were subsequently performed. Machine learning models were then constructed, and SHapley Additive exPlanations (SHAP) analysis was used to prioritize key genes. Finally, the selected gene was experimentally validated using clinical liver tissue samples, free fatty acid (FFA)-induced steatosis models in AML12 and HepG2 cells, qRT-PCR, Western blotting, and Oil Red O staining. Ten KMRGs closely associated with MAFLD were identified. Functional enrichment analysis showed that these genes were mainly involved in PPAR signaling, epigenetic regulation, and FOXO-mediated oxidative stress and metabolic transcriptional processes. A transcriptome-based classification model for MAFLD was constructed using the Naive Bayes machine learning algorithm, and SHAP analysis identified SCO2 as a feature with relatively high importance in model prediction. External validation demonstrated that KMRGs exhibited abnormal expression patterns, diagnostic potential, and correlations with NAFLD activity score (NAS) across different cohorts. Further analyses showed that KMRGs were closely associated with immune microenvironment remodeling, inflammatory responses, lipid metabolic dysregulation, and fibrosis in MAFLD, and could define molecular subtypes with distinct pathological features. Experimental validation showed that SCO2 was highly expressed in clinical MAFLD samples and in FFA-induced in vitro steatosis models. Functional assays further indicated that SCO2 modulated FFA-induced lipid accumulation in vitro. KMRGs are closely associated with the presence and pathological activity of MAFLD. Among them, SCO2 showed high feature importance in machine learning analyses and was associated with FFA-induced lipid accumulation in vitro, suggesting that it is a candidate molecule for further mechanistic investigation in MAFLD.