Interpretable Screening of Illicit Drugs in Raw Urine via LightGBM-Assisted SERS

Abstract Illicit drug-related offenses are a rapidly escalating global concern, driving an urgent need for rapid and reliable drug screening systems. Although conventional identification methods provide definitive results, they often require prolonged turnaround times and complex procedures, limiting their utility in point-of-care (POC) applications. Here, we report a roughened gold nanograss (RGNG) surface-enhanced Raman scattering (SERS) substrate designed for sensitive and automated narcotics screening. The hierarchical RGNG architecture, optimized through Au–Ag codeposition followed by selective etching, generates high-density plasmonic hotspots and achieves high analytical sensitivity, even in complex biological matrices. In vivo studies using rat models demonstrated successful classification of four illicit drugs in urine for up to 6 h after administration using principal component analysis. Furthermore, integration with an explainable LightGBM machine learning framework enabled differentiation (AUC = 0.80) of illicit drug-positive samples from heavily medicated negative-control samples across 155 clinical urine specimens by distinguishing drug-associated biomarker patterns from systemic metabolic backgrounds. This diagnostic pipeline, which combines advanced plasmonic engineering with interpretable gradient-boosting algorithms, provides a scalable proof-of-concept approach for rapid forensic and medical screening; external validation in an independent cohort will be required before clinical deployment.