Enhanced reflectance-based terahertz biosensing using tunable graphene metasurfaces for label-free peptide detection: a machine learning-optimized platform for biomedical diagnostics

We propose a reflectance-based terahertz (THz) metasurface biosensor that integrates tunable graphene components for highly sensitive, label-free peptide detection in biomedical applications. Through COMSOL Multiphysics simulations employing the finite element method, we demonstrate outstanding sensor performance, achieving a peak sensitivity of 0.279 THz/RIU, a figure of merit of 15.5 RIU−1, a quality factor above 50, and a detection limit of 0.048 RIU across the 0.1–0.45 THz frequency range. The sensor exhibits excellent angular stability, with reflectance increasing from 66.251% to 91.305% for incidence angles between 0° and 80°. By tuning graphene’s chemical potential (0.1–0.9 eV), dynamic spectral control is achieved, enhancing reflectance from 14.947% to 70.919% – a performance surpassing that of conventional absorptance-based sensors. Parametric optimization reveals key geometric dependencies, identifying optimal resonator dimensions for maximum performance. Furthermore, machine learning – assisted optimization using Gradient Boosting Regression attains prediction accuracies above 90% for both refractive index variations and angular responses.