Molecular enrichment integrated with soft voting for efficient SERS detection of multiple drugs

Abstract

Background

Surface-enhanced Raman spectroscopy (SERS) is regarded as a powerful tool for rapid drug identification in the liquid phase. However, the implementation of SERS for drug detection in the liquid phase remains constrained by two critical limitations: inadequate molecular enrichment efficiency at hotspots and spectral interpretation difficulties arising from complex multi-analyte interactions. Besides, manual scrutiny of large-scale SERS datasets is painstakingly slow and overwhelmingly labor-intensive. Currently, it is clear that an effective molecular enrichment strategy combined with machine learning models with high accuracy are research hotspots. (85)

Results

We designed a novel AuNPs@AgCl (Au nanoparticles (AuNPs) modified with activated sealing layer of AgCl) substrate through the reaction of Cl^- ions and Ag⁺ ions to address these issues. This substrate achieved ultralow detection limits for diverse substances: 4 × 10⁻¹¹ M for crystal violet, 0.2 ppm for morphine, 12.5 ppm for methamphetamine, and 10 ppm for methadone. Meanwhile, the results showed that the AuNPs@AgCl substrate was able to quantify methadone spiked in lake water, artificial saliva and artificial blood with a lower limit of detection of 50 ppb, 0.8 ppm, and 0.6 ppm, respectively. Furthermore, we developed a soft voting classifier which incorporates random forest (RF), extra-trees (ET), XGBoost and support vector machine (SVM) models. This architecture delivers 90.7% average accuracy in classifying 6 drugs combinations (unary/binary mixtures), outperforming single-algorithm approaches by 21.4-26.7%. (134)

Significance

This rationally designed substrate effectively improved the probability of analytes falling into hotspots and enhanced SERS stability. Besides, the soft voting classifiers have demonstrated their great potential to improve the classification accuracy for both unary and binary drugs in a balanced manner. This synergistic combination of molecular enrichment and soft voting classifier holds great potential for drug detection applications. (59)