Integrative Transcriptomic Analysis and Interpretable Machine Learning Uncover a Core Biomarker Signature and Its Immu-noregulatory Network Mediating the Sodium Arsenite-MASLD Axis

Abstract Sodium arsenite, a ubiquitous environmental hepatotoxin, has an ill-defined role in the pathophysiology of metabolic dysfunction-associated steatotic liver disease (MASLD). This study aimed to systematically identify key molecular targets linking sodium arsenite exposure to MASLD and to elucidate their potential mechanisms in regulating the liver immune microenvironment. Leveraging the GSE135251 transcriptomic dataset, we employed a dual-strategy approach combining Weighted Gene Co-expression Network Analysis (WGCNA) and Differential Expression Analysis (DEA) to identify high-confidence genes associated with MASLD. An intersection with network toxicology databases yielded 283 core candidate genes. To distill the most predictive biomarkers, we deployed an ensemble machine learning pipeline featuring LASSO, RF-RFE, and Random Forest. We identified a robust 11-gene biomarker signature, which was significantly dysregulated in MASLD patients and formed a highly synergistic internal regulatory network. Immune infiltration analysis demonstrated a significant increase in both M1 and M2 macrophages in MASLD livers. Crucially, the expression of the 11 core genes strongly correlated with macrophage subset abundances (e.g., DUSP1 was strongly negatively correlated with M1 macrophages, r=-0.70). Functional enrichment highlighted inflammatory pathways, including TNF and IL-17 signaling. Finally, drug repositioning identified 14 approved drugs targeting FOS, KDM4A, and DUSP1. Through an innovative, multi-layered analytical framework, this study successfully identified a sodium arsenite-associated 11-gene signature for MASLD, providing profound insights into its immunomodulatory role and presenting novel therapeutic targets.