<b>Study on blood diagnostic markers of diabetic nephropathy based on </b><b>metabolomics</b><b> and machine learning</b>

<b>Aim: </b>To identify potential biomarkers and explore the mechanisms underlying diabetic nephropathy (DN) by non-targeted metabolomics studies and machine learning algorithms.<b>Materials and methods: </b>In this study, blood samples were collected from 51 patients with type 2 diabetes mellitus (which will be divided into 21 in the normal albuminuria (NA) group, 13 in the microalbuminuria (MI) group, 11 in the massive albuminuria (MA) group and 6 in the Kidney Failure (KF) group based on the amounts of urinary proteins and the diagnostic record on admission) as well as 26 healthy subjects, and then performed metabolomics experiments on them.<b>Results: </b>The results showed that pyrimidine metabolism, fructose and mannose metabolism, steroid hormone biosynthesis were closely related to the progression of DN. Three metabolites (sorbitol, adipic acid, and N-Ribosylhistidine) were finally identified as potential biomarkers by machine learning algorithms, and all of them were positively correlated with urinary albumin creatinine ratio (UACR) and serum creatinine (SCr).<b>Conclusions: </b>This study successfully screened three differential metabolites as a set of potential DN biomarkers and estimated them to have good predictive performance by receiver operating characteristic curve (ROC) modeling, which provides new perspectives for early diagnosis and probing of disease mechanisms of DN.