BACKGROUND

Diabetic nephropathy (DN) is a complication of systemic microvascular disease in diabetes mellitus. Abnormal glycolysis has emerged as a potential factor for chronic renal dysfunction in DN. The current lack of reliable predictive biomarkers hinders early diagnosis and personalized therapy.

METHODS

Transcriptomic profiles of DN samples and controls were extracted from GEO databases. Differentially expressed genes (DEGs) and their functional enrichments were identified. Glycolysis-related genes (GRGs) were selected by combining DEGs, weighted gene co-expression network, and glycolysis candidate genes. We established a diagnostic signature termed GScore via integrative machine learning framework. The diagnostic efficacy was evaluated by decision curve and calibration curve. Single-cell RNA sequence data was used to identify cell subtypes and interactive signals. The cMAP database was used to find potential therapeutic agents targeting GScore for DN. The expression levels of diagnostic signatures were verified .

RESULTS

Through the 108 combinations of machine learning algorithms, we selected 12 diagnostic signatures, including CD163, CYBB, ELF3, FCN1, PROM1, GPR65, LCN2, LTF, S100A4, SOX4, TGFB1 and TNFAIP8. Based on them, an integrative model named GScore was established for predicting DN onset and stratifying clinical risk. We observed distinct biological characteristics and immunological microenvironment states between the high-risk and low-risk groups. GScore was significantly associated with neutrophils and non-classical monocytes. Potential agents including esmolol, estradiol, ganciclovir, and felbamate, targeting the 12 diagnostic signatures were identified. , ELF3, LCN2 and CD163 were induced in high glucose-induced HK-2 cell lines.

CONCLUSION

An integrative machine learning frame established a novel diagnostic signature using glycolysis-related genes. This study provides a new direction for the early diagnosis and treatment of DN.