Identification of hub genes involved in the pathogenesis of diabetic nephropathy: A multi‐omics study integrating machine learning, mendelian randomization and mediation analysis

Abstract Background Diabetic nephropathy (DN), affecting 30%–40% of diabetic patients, is the leading cause of end‐stage renal disease worldwide. This study aims to identify diagnostic biomarkers and explore potential gene‐metabolite interactions in DN pathogenesis through integrated bioinformatics approaches and experimental validation. Methods We analysed Gene Expression Omnibus datasets through differential expression analysis, weighted gene co‐expression network analysis (WGCNA) and machine learning algorithms to identify key DN‐associated genes. The causal relationships between candidate genes and DN were investigated using Mendelian randomization (MR) analysis with cis‐expression quantitative trait loci data, which random‐effects inverse variance weighted (IVW) method was our primary analysis approach. We then employed mediating MR analysis to explore metabolite‐mediated pathways. The expression of identified genes was validated in both human DN kidney samples and diabetic mouse models. Results Our analysis identified seven diagnostic genes (LUM, ESM1, VCAN, LOX, G0S2, THBS2 and UMOD). Through MR analysis, THBS2 showed a significant causal association with DN outcomes (OR = 1.1653, 95% CI [1.0103, 1.3441], p = 0.0357). Subsequent mediation analysis suggested that THBS2 may influence DN development through modulation of N‐acetyl‐beta‐alanine levels (mediating effect = 17.85%, p = 0.02). Importantly, elevated THBS2 expression was confirmed in both DN patient kidney samples and diabetic mouse models, validating our bioinformatics findings. Conclusions Through comprehensive bioinformatics analysis and experimental validation, we identified THBS2 as a potential diagnostic marker for DN. These findings provide new insights into DN pathogenesis and suggest directions for future therapeutic strategies.