Proteoliposomes on 2D-MoS₂ plasmonic nanocavities for enhanced Raman spectroscopy with machine learning-based identification and classification

Synthetic proteoliposomes functionalized with disease-relevant surface markers offer a powerful platform for modeling biological vesicles such as lipid nanoparticles and extracellular vesicles. The simplified composition of proteoliposomes facilitates the design and interpretation of analytic approaches for classifying vesicles and characterizing their contents. Here, we present a library of synthetic proteoliposomes incorporating tumor-associated surface biomarkers—EGFR, α6β4, and αvβ5—and nucleic acid cargo, to mimic cancer-derived extracellular vesicle phenotypes. For molecular fingerprinting, we employed a custom-designed 2D-plasmonic nanocavity platform that enables high-resolution, label-free Surface-Enhanced Raman Spectroscopy (SERS). Integrated with supervised machine learning algorithms, including Random Forest Classifier (RFC) and Support Vector Machine (SVM), this system achieved robust classification of proteoliposome subtypes with test accuracies of 82% and 76%, respectively. Our results demonstrate the power of combining synthetic vesicle engineering with advanced optical sensing for capturing subtle biomolecular differences. This platform enables standardized, interpretable diagnostic readouts and offers a versatile tool for probing molecular interactions in lipid-based systems such as virus-like particles and nanotherapeutics.