Purpose/Significance To identify multi-gene biomarkers of diabetic nephropathy(DN)，and to construct an interpretable prediction model，so as to provide references for early diagnosis of the disease. Method/Process Based on the GEO database， GSE111154 and GSE96804 are used as the training sets，and GSE30122 as the validation set. Key genes are screened using random for-est(RF)and least absolute shrinkage and selection operator(LASSO). Three machine learning algorithms are employed to constructpredictive models and evaluate their performance. Functional enrichment and SHAP interpretability analysis are conducted by combininggene ontology(GO)with Kyoto encyclopedia of genes and genomes(KEGG). Result/Conclusion Differentially expressed genes are en-riched in amino acid metabolic reprogramming，complement and coagulation cascade activation，extracellular matrix remodeling and RAGE signaling pathway，and may be involved in disease progression through the“immune-metabolic-fibrosis”axis. FN1，NT5E and TPM1 are identified as key genes. SHAP analysis reveal that TPM1 and FN1 genes are the core predictors.