Integrating bioinformatics and machine learning to elucidate the role of protein glycosylation-related genes in the pathogenesis of diabetic kidney disease

Abstract

Background Diabetic kidney disease (DKD) is a severe global complication of diabetes, yet its molecular mechanisms remain incompletely understood. This study aimed to investigate the role of protein glycosylation in DKD pathogenesis and its association with gene expression changes, with the goal of identifying diagnostic biomarkers and personalized therapeutic targets. Methods Integrated bioinformatics and machine learning approaches were applied to analyze multiple gene expression datasets. Differentially expressed glycosylation-related genes were identified, followed by unsupervised clustering to define molecular subtypes. Functional enrichment, immune cell infiltration analysis, and machine learning algorithms (including feature selection for hub genes) were employed. qPCR validation was performed on clinical DKD and normal kidney tissues, and ROC curves were generated to assess diagnostic potential. Results Unsupervised clustering of glycosylation-related genes revealed two distinct DKD molecular subtypes with differential pathway activation (e.g., extracellular matrix remodeling) and immune infiltration patterns. Six hub genes (S100A12, EXT1, SBSPON, ADAMTS1, FMOD, SPTB) were identified as critical to DKD pathogenesis through machine learning. Immune infiltration analysis showed significant differences in macrophage and neutrophil activity between DKD and controls and Immunohistochemical results confirmed the occurrence of immune infiltration. qPCR validation confirmed dysregulation of hub genes in DKD tissues compared to normal samples. ROC analysis demonstrated high diagnostic accuracy for these genes. Conclusions This study highlights abnormal protein glycosylation as a key player in DKD and identifies six hub genes with potential as diagnostic biomarkers. The molecular subtypes and immune infiltration patterns provide insights into disease heterogeneity, paving the way for personalized therapies. Future studies should validate these findings in larger cohorts with explicit sample sizes to strengthen clinical applicability.