Source Tracing of Kidney Injury via the Multispectral Fingerprint Identified by Machine Learning-Driven Surface-Enhanced Raman Spectroscopic Analysis

Abstract

Early diagnosis of drug-induced kidney injury (DIKI) is essential for clinical treatment and intervention. However, developing a reliable method to trace kidney injury origins through retrospective studies remains a challenge. In this study, we designed ordered fried-bun-shaped Au nanocone arrays (FBS NCAs) to create microarray chips as a surface-enhanced Raman scattering (SERS) analysis platform. Subsequently, the principal component analysis (PCA)-two-layer nearest neighbor (TLNN) model was constructed to identify and analyze the SERS spectra of exosomes from renal injury induced by cisplatin and gentamycin. The established PCA-TLNN model successfully differentiated the SERS spectra of exosomes from renal injury at different stages and causes, capturing the most significant spectral features for distinguishing these variations. For the SERS spectra of exosomes from renal injury at different induction times, the accuracy of PCA-TLNN reached 97.8% (cisplatin) and 93.3% (gentamicin). For the SERS spectra of exosomes from renal injury caused by different agents, the accuracy of PCA-TLNN reached 100% (7 days) and 96.7% (14 days). This study demonstrates that the combination of label-free exosome SERS and machine learning could serve as an innovative strategy for medical diagnosis and therapeutic intervention.