Rapid Surface-Enhanced\nRaman Scattering Imaging and\nDeep Learning for Highly Sensitive Discrimination of Amino Acids and\nPeptides

Developing a highly sensitive and\naccurate method to\ndiscriminate\nbetween amino acids and peptides is vital for establishing future\nhealthcare testing technologies, such as liquid biopsy. This study\nproposes a highly sensitive technique based on surface-enhanced Raman\nscattering (SERS), which combines chemically linking an analyte with\ngold nanoparticles and aggregating them to produce hotspots. Furthermore,\nby combining rapid SERS imaging with slit-scanning Raman microscopy\nand deep learning based on a convolutional neural network, 20 proteinogenic\namino acids were successfully detected and distinguished with accuracies\nexceeding 95%. Also, out of 39 types of dipeptides that have Phe at\neither the amino terminal or the carboxyl terminal, 19 types were\nidentified with high accuracy. Even for dipeptides with lower identification\naccuracy, it was confirmed that they were recognized as one of the\ndipeptides with high structural similarity, such as cyclic structures\nand branched amino acids. Moreover, pathophysiologically relevant\nsequence differences in β-amyloid peptides were accurately discriminated\nwith a sensitivity of approximately 975 zeptomoles.