Discrimination of self-assembled monolayer composition on gold nanostructure surface by combining machine learning and surface enhanced Raman scattering

Raman spectroscopy is a spectroscopic method which indirectly measures molecular vibration and can identify molecules by obtaining their unique vibrational fingerprints. Raman spectroscopy has the advantage of non-invasive and label-free detection, but it has the problem of low sensitivity due to the very weak intensity of Raman scattered light. In order to make Raman spectroscopy high sensitivity, surface enhanced Raman scattering (SERS) induced via metallic nanostructures has been widely studied. SERS is attracting attention in various fields of analysis such as bioanalysis, material analysis, environmental analysis, and food analysis because of its high sensitivity and selectivity. However, SERS spectra of adsorbed molecules with similar structures have very similar shapes, making it difficult to discriminate molecules from their spectra. Especially, it is even more difficult to determine the mixing ratio of mixtures of those molecules. Thus, we focused on the use of machine learning to determine the mixing ratio of similar structural molecules on gold nanostructure. In this study, we fabricated gold nanostructures by depositing gold on a cyclo-olefin polymer (COP) film with periodical nanostructure. Then, self-assembled monolayer (SAM) of mixed benzene thiol derivatives, as a model of surface adsorbed molecules, were prepared on the surface of gold nanostructure, and measured SERS spectra. We examined several machine learning models that can accurately determine mixing ratios from the obtained spectra. As a result, we succeeded in determining the mixing ratios of molecules with approximately 99% accuracy, with multilayer perceptron (MLP) model being the most accurate.