Molecule-Responsive SERS Sensors for Urine Diagnosis of Kidney Diseases Enhanced by Neural Networks

Early diagnosis of kidney disease is crucial for treatment and prognosis. Compared with kidney biopsy, a noninvasive urine-based diagnosis method of kidney disease can be more convenient and less painful for patients. Urine is closely associated with kidney disease, including nephritis, kidney failure, and kidney cancer. Since it contains various biomolecules, including small-molecule metabolites, proteins, and so on, urine is an appropriate sample for diagnosing and monitoring kidney disease progression. Herein, we developed a liquid biopsy method for diagnosing various kidney diseases based on a specific SERS sensing mode combined with neural network models, which allows an integrated response to multiple types of targets with a single probe and facilitates the detection of complex samples from multiple target groups. Compared to label-free SERS, this method relies on changes in the probe molecule, which facilitates the sensitivity of the assay. Two kinds of silver nanoparticle-casting films assist this method with the surface decoration of molecule-responsive Raman reporter molecules: p-mercaptobenzoic acid (MBA) and p-aminothiophenol (PATP). MBA responds to amino components in urine by SERS spectral changes caused by molecular polarizability. PATP illustrates the level of small-molecule metabolites in urine according to SERS changes resulting from the rate of the hot hole-catalyzed reactions. These interactions were evidenced by density functional theory and molecular docking simulations. Through these two SERS sensors, we acquired the SERS data sets of urine samples and established a classifier by incorporating neural network models, enabling the effective discrimination of healthy and kidney disease samples. The method is helpful for clinical validation and shows promise for use in long-term kidney disease monitoring programs.