Interpretable Machine Learning Identifies Hub Biomarkers of Renal Fibrosis and Their Potential Medical Applications.

BACKGROUND: Renal fibrosis is a crucial pathogenic driver of chronic kidney disease (CKD). However, its heterogeneous limits accurate assessment by renal biopsy. The study aimed to identify accurate diagnostic biomarkers and potential therapeutic targets for renal fibrosis. METHODS: We analyzed renal fibrosis transcriptomic datasets from the GEO database to identify differentially expressed genes (DEGs). Hub genes were selected through the Least Absolute Shrinkage and Selection Operator (LASSO) regression, with their association to immune infiltration subsequently analyzed using CIBERSORT. Interpretable machine learning models, specifically eXtreme Gradient Boosting (XGBoost) and Deep Neural Network (DNN), were developed for sample classification, with their interpretability and key biomarker contribution assessed through Shapley Additive Explanations (SHAP) analysis. The predicted hub genes were validated using histological staining, Western blot (WB) experiments, and functional cellular assays in rat renal fibroblast cells and mouse renal fibrosis models. Finally, potential therapeutic drugs targeting the hub genes were identified through molecular docking. RESULTS: We identified 26 fibrosis-related genes for renal fibrosis and established their correlations with inflammatory and immune infiltration. Machine learning models demonstrated high diagnostic accuracy (XGBoost: 96%; DNN:92%). SHAP analysis highlighted AGR2 and DOCK2 as top predictors. Subsequent experimental validation confirmed their significant upregulation and functional involvement in fibrotic processes. Molecular docking identified several existing drugs such as Dexamethasone and Ciclosporin as potential AGR2-targeting agents. CONCLUSION: This study identifies AGR2 and DOCK2 as novel biomarkers and therapeutic targets for renal fibrosis, highlighting their dual potential for diagnostic application and targeted therapy development.