Abstract
We report a proof-of-concept diagnostic strategy that integrates multiplexed Raman-tagged antibody labeling with label-free surface-enhanced Raman spectroscopy (SERS) and machine learning (ML) to improve the detection of ovarian cancer via extracellular vesicles (EVs). EVs were isolated from patient plasma using size-exclusion chromatography and labeled with polyyne-based Raman tags targeting three ovarian cancer biomarkers: CA-125, HE4, and CA-19-9. Labeled and unlabeled EVs were deposited onto SERS-active substrates, and spectra were collected using a custom confocal Raman microscope. Incorporating the tag-derived signal into SERS analysis enhanced interpretability and added molecular specificity. We evaluated classification performance using various ML models applied to spectral datasets from a cohort of ovarian cancer patients and healthy controls. Combined use of the Raman tag and label-free regions improved classification accuracy compared to either modality alone. Notably, support vector machine (SVM) achieved over 95 % accuracy, sensitivity, and specificity. Compared to ELISA, our SERS platform demonstrated improved sensitivity in detecting EV-associated biomarkers from small sample volumes. This approach addresses a key limitation of SERS-based diagnostics by linking spectral features to known biomarkers, offering improved transparency and performance in ML-enabled liquid biopsy.