Application of proteomics and machine learning algorithms to investigate the mechanisms of diabetic nephropathy and screen urinary biomarkers

Diabetic nephropathy (DN) has become the main cause of end-stage renal disease worldwide, bringing serious harm. Early diagnosis of the disease is quite inadequate. To screen urine biomarkers of DN and explore its potential mechanism, this study collected urine from 87 patients with type 2 diabetes mellitus (T2DM) (which will be classified into normal albuminuria (NA), microalbuminuria (MI), macroalbuminuria (MA) groups based on the amount of urinary protein) and 38 healthy subjects and randomly chose 12 from each group as a screening cohort and the rest as a validation cohort. The results showed that humoral immune response, complement activation, complement and coagulation cascade, renin-angiotensin system (RAS) and cell adhesion molecules were closely related to the progression of DN. Five overlapping proteins (KLK1, CSPG4, PLAU, SERPINA3, and ALB) were identified as potential biomarkers by machine learning methods. Among them, KLK1 and CSPG4 were positively correlated with urinary albumin to creatinine ratio (UACR), and SERPINA3 was negatively correlated with UACR, which were validated by enzyme-linked immunosorbent assay (ELISA) experiments. This study provides new insights into disease mechanisms and biomarkers for early diagnosis of DN.