Machine learning enhanced terahertz metasurface biosensor based on MXene–Graphene–Black phosphorus heterostructure for peptide detection

Terahertz metasurface biosensors offer a promising route for label-free and non-invasive biomolecular detection; however, many existing designs rely on limited material platforms and operate over narrow refractive-index ranges, restricting their practical applicability in complex biological environments. In this work, a hybrid MXene–graphene–black phosphorus (BP) based terahertz metasurface biosensor is proposed for the quantitative detection of peptide biomolecules. The sensor is systematically analyzed using finite element method simulations in COMSOL Multiphysics to optimize the resonator geometry and investigate its electromagnetic response. The influence of key parameters, including graphene Fermi level, incident wave angle, and structural dimensions, on the transmission spectra is comprehensively examined. The optimized design achieves a high refractive-index sensitivity of 500 GHz/RIU over a broad operating range of 1.606–1.706 RIU, which is well suited for peptide sensing applications. To accelerate performance evaluation, a machine-learning framework based on Bayesian Ridge Regression is employed to predict resonance frequency shifts and transmission characteristics, yielding a strong predictive accuracy with R 2 = 0.94. Comparative analysis confirms the competitive sensitivity and wide dynamic range of the proposed sensor. Overall, the results demonstrate the strong potential of the proposed metasurface platform for rapid, label-free, and quantitative peptide biomarker detection in terahertz diagnostic applications.