Several fabrication methods have been developed for label-free detection in various fields. However, fabricating high-density and highly ordered nanoscale architectures by using soluble processes remains a challenge. Herein, we report a biosensing platform that integrates deep learning with surface-enhanced Raman scattering (SERS), featuring large-area, close-packed three-dimensional (3D) architectures of molybdenum disulfide (MoS2)-assisted gold nanoparticles (AuNPs) for the on-site screening of coronavirus disease (COVID-19) using human tears. Some AuNPs are spontaneously synthesized without a reducing agent because the electrons induced on the semiconductor surface reduce gold ions when the Fermi level of MoS2 and the gold electrolyte reach equilibrium. With the addition of polyvinylpyrrolidone, a two-dimensional large-area MoS2 layer assisted in the formation of close-packed 3D multistacked AuNP structures, resembling electroless plating. This platform, with a convolutional neural network-based deep learning model, achieved outstanding SERS performance at subterascale levels despite the microlevel irradiation power and millisecond-level acquisition time and accurately assessed susceptibility to COVID-19. These results suggest that our platform has the potential for rapid, low-damage, and high-throughput label-free detection of exceedingly low analyte concentrations.
Keywords: energy equilibrium state (EF and Eredox); molybdenum disulfide; multistacked gold nanoparticles; surface-enhanced Raman scattering; tear fluids with coronavirus.