Multi-omics and machine learning identify FN1 and ALDH2 as diagnostic biomarkers and therapeutic targets in early and late diabetic kidney disease

Diabetic kidney disease (DKD), the leading cause of end-stage kidney disease worldwide, demands deeper molecular characterization to improve clinical management. This study employed an integrated multi-omics approach to identify stage-specific biomarkers and molecular mechanisms distinguishing early- and late-stage DKD. Initial bulk RNA-seq analysis (thresholds: |logFC|>0.585, FDR < 0.05) revealed differentially expressed genes and enriched pathways, followed by two-sample Mendelian randomization (IVW p < 0.05) to pinpoint causal genes. Diagnostic modeling combined LASSO regression (10-fold cross-validation) with four machine learning algorithms (Random Forest, SVM, GLM, and XGBoost), validated in an independent cohort. Single-cell resolution analysis mapped candidate gene expression patterns, while molecular docking screened potential therapeutics. Through an integrated validation pipeline, FN1 (OR = 1.32) and ALDH2 (OR = 0.76) were established as core diagnostic biomarkers. FN1 (upregulated in mesangial cells) and ALDH2 (downregulated in proximal tubules) were validated as stage-specific biomarkers of DKD progression. FN1 expression negatively correlated with eGFR decline, while ALDH2 positively correlated. Resveratrol showed high-affinity docking to FN1 and ALDH2, suggesting dual-target therapeutic potential. These findings position FN1, ALDH2, and resveratrol as central references for DKD biomarker discovery, progression staging, and therapeutic development.