Machine-Learning-Enabled Microfluidic SERS With Charge-Directed 3D Au@Ag Plasmonic Assemblies for Therapeutic Drug...

Point-of-care compatible therapeutic drug monitoring requires platforms to rapidly quantify low-abundance small-molecule drugs in complex biofluids while resolving spectral interference from coexisting compounds. Surface-enhanced Raman spectroscopy offers molecular fingerprint specificity, yet its quantitative reliability is often compromised by heterogeneous hotspots, matrix interference, and overlapping drug spectra. Here, we develop an integrated microfluidic SERS platform based on charge-directed 3D Au@Ag plasmonic assemblies and machine learning-assisted spectral decoding. Oppositely charged gold nanopolytopes and silver nanospheres spontaneously assemble into ordered 3D structures with dense interparticle nanogaps. Optimized 1:1 Au@Ag assembly delivers enhanced electromagnetic coupling and stable, reproducible SERS signals. Combined with deuterated methanol as a ratiometric internal standard in microfluidic chips, this system achieves low-volume, fluctuation-corrected standardized drug analysis. Six therapeutic molecules are identified via SERS fingerprints, and binary or ternary mixtures are accurately decoded by 3D-LDA and CNN-RF with AUC > 0.98. Reliable discrimination in rat plasma, artificial sweat, and urine demonstrates excellent matrix tolerance. This work establishes a material microfluidic algorithm-integrated SERS strategy for rapid therapeutic drug analysis, paving the way toward portable intelligent drug monitoring systems.