Identification of diagnostic biomarkers for diabetes nephropathy by multi-chip integrated bioinformatics combining machine-learning strategies and mendelian randomization

Abstract

Background Diabetic nephropathy (DN) represents a significant complication of diabetes, progressively contributing to the global incidence of end-stage renal disease (ESRD). Currently, the diagnosis of DN primarily relies on traditional clinical indicators, which may not adequately reflect the progression and prognosis of all DN patients, posing challenges in developing diagnostic and therapeutic strategies. This study aims to identify new molecular and genetic biomarkers closely associated with the onset and progression of DN through bioinformatics and genetic epidemiology analysis methods. Methods Gene expression profiles were obtained from gene expression omnibus (GEO) database. The R software was employed to filter differentially expressed genes (DEGs) and to conduct enrichment analyses of these genes. It is remarkable that five distinct machine learning classifiers were utilized to identify diagnostic biomarkers and to formulate a diagnostic model. These biomarkers were further validated in an external validation dataset generated by Nephroseq V5, whereupon a clinical characteristic correlation analysis ensued. In light of the machine learning results, immune infiltration analysis and single-cell RNA sequencing were undertaken. Ultimately, the mendelian randomization method was used to examine the causal relationship between the identified biomarkers and DN. Results Five biomarkers implicated in DN, including AFM, DUSP1, KRT19, TGFBI, and ZFP36, were subjected to external testing and validation, utilizing various machine learning models with high diagnostic efficacy. Correlation analysis demonstrated that the expression of these biomarkers correlated with the deterioration of kidney function, and single-cell RNA sequencing results indicated that the biomarkers were predominantly localized in granulocytes, macrophages, and monocytes. The results from the inverse variance weighted (IVW) analysis elucidated that DUSP1 (OR = 0.664) serves as a protective factor for DN, whereas TGFBI (OR = 1.114) constitutes a risk factor for DN, in agreement with bioinformatics analysis. Conclusions The present study may provide new insights into the mechanisms underlying DN onset and progression, as well as the selection of DN diagnostic markers and therapeutic targets.