Single-Molecule Peptide Discrimination via Flow-Through SERS and Machine Learning

Peptides are key biomolecules in biology and medicine, yet their reliable detection and discrimination in complex mixtures remain highly challenging, particularly under clinical requirements of robustness and accuracy. Surface-enhanced Raman spectroscopy (SERS) offers molecular specificity but is hindered by spectral overlap, variability, and fluctuations that limit its applicability in practical settings. In this work, we investigate the performance of SERS flow-through strategy using plasmonic nanopores to record Raman spectra from single molecules as they translocate one by one through sub-2 nm hotspots. As a stringent test, we investigated the discrimination of vasopressin and oxytocin, two peptides differing by only two amino acids. Using electrophoretic delivery and ultrafast SERS detection with a single-photon avalanche diode camera, we captured spectra on microsecond time scales. Machine-learning analysis achieved 70.5% classification accuracy at the single-peptide level, rising to 99% discrimination when averaging 40 events. These results establish flow-through nanopore SERS as a promising route toward single-molecule peptide identification in biomedical settings.