Deep Learning-Enabled Interpretative SERS Analysis of Key Urinary Metabolites in Nephropathy Using Interfacial Charge-Engineered Silver Nanoparticles

Abstract

The definitive diagnosis of kidney disease relies on the invasive gold-standard method of kidney biopsy, but this provides information only about local characteristics. In contrast, urine samples can be collected noninvasively and contain a variety of metabolites associated with kidney disease. However, obtaining high-quality, high-dimensional biological information and identifying specific diagnostic biomarkers in urine also pose significant challenges. This study developed a detection strategy that uses interface charge engineering with patient urine samples to diagnose diabetic nephropathy, IgA nephropathy, lupus nephritis, and nephrotic syndrome. Different sample types form biomolecular crowns of varying thicknesses at the detection interfaces. Surface-enhanced Raman scattering (SERS) technology was employed to capture the urine metabolite fingerprints. We established a multiclassification model by combining urine spectral data with deep learning, conducted an interpretive analysis to clarify the basis for classification, and then used enrichment analysis to explore urinary metabolic information. This study establishes a highly sensitive, noninvasive, rapid, cost-effective, and patient-friendly classification method for kidney diseases, which is valuable for precision medicine.