Dendritic Ag@c-Si structure for influenza A virus detection by SERS and machine learning

Abstract

The combination of Surface-enhanced Raman Scattering (SERS) method and machine learning algorithms has been successfully demonstrated for the detection and differentiation of influenza A virus in buffer medium. Dendritic Ag nanostructures on single-crystal Si wafer have been proposed as SERS-active substrates, which are fabricated using a low-cost and reproducible AgNO3 reduction method. Depending on the time of Ag deposition, structures of different morphologies were obtained. These structures were studied by spectroscopic ellipsometry (SE). To analyze the spectra obtained by SE, we used the Bruggeman effective medium approximation to determine the volume fraction and height of the metal, the Drude model to describe free electrons, the Tauc-Lorentz model to describe the influence of interband transitions, as well as Lorentz model and Gauss model to describe localized surface plasmon resonance (LSPR). SERS was achieved through LSPR excitation and the presence of “hot spots” in the regions around the tips and in the spaces between close-packed dendritic Ag nanostructures. The sample with the most developed surface is the most promising for SERS, since the spectral position of the LSPR is at 658 nm, which is close to the excitation wavelength of the He–Ne laser. The SERS spectra of influenza A virus in buffer medium are difficult to distinguish, so a machine learning algorithms (principal component analysis and support vector machine) were used to directly classify them. For the sample with the most developed morphology, the detection accuracy of influenza A virus was 76.6 ± 4.2 % at a total protein concentration in the test analyte of 300 μg/ml (according to the Lowry method).