Identification of key biomarkers for ferroptosis in diabetic kidney disease using machine learning and WGCNA.

OBJECTIVE: Early detection of diabetic kidney disease (DKD) remains challenging because currently available clinical markers mainly reflect established renal injury rather than early pathogenic changes. Ferroptosis has been increasingly implicated in DKD development, we aimed to identify ferroptosis-related molecular signatures and candidate diagnostic biomarkers for DKD. METHODS: We integrated public kidney transcriptomic datasets from DKD and control samples and analyzed them together with curated ferroptosis-related genes. Differential expression and gene co-expression analyses were combined with protein-interaction and machine-learning approaches to prioritize candidate biomarkers. Single-cell transcriptomic analysis was used to explore cellular localization, and the leading candidate was further validated in an independent clinical transcriptomic database and a diabetic mouse model. RESULTS: We identified genes that were consistently dysregulated across DKD datasets and linked them to ferroptosis-related processes. Network-based analysis prioritized several hub genes, among which CD36 emerged as the most robust candidate biomarker. Independent database validation demonstrated that CD36 was significantly upregulated and associated with renal function indicators. Consistent with these findings, experimental validation confirmed increased renal CD36 protein expression and elevated serum CD36 levels in diabetic mice, further supporting its potential as a biomarker for DKD. CONCLUSIONS: Our ferroptosis-oriented screening strategy identified CD36 as a promising diagnostic biomarker for DKD. These findings support the potential value of ferroptosis-related signatures for biomarker discovery in DKD and warrant further mechanistic validation.