A Hybrid Machine Learning and Analytical Model of Biofets for the Detection of Peptides

BioFETs enable the detection of a diverse range of biological analytes by transducing changes in surface potential into a measurable electrical signal. Although established analytical models, including the Gouy-Chapman-Stern and site-binding frameworks, effectively describe surface charge and protonation equilibria, they do not account for the sequence-specific chemistry, steric effects, or conformational diversity characteristic of peptides and other biomolecules. Here we present a hybrid model to address these limitations, augmenting previous BioFET simulation frameworks with molecular parameters including proton dissociation constants and topological surface areas determined through molecular modelling. While this work focuses on peptides, our framework extends to modified peptides, diverse biomolecular analytes and immobilization chemistries relevant to biosensing. We validate our model against experimental values and molecular modelling studies, confirming the physical fidelity of our hybrid model and perform simulations of representative peptides demonstrating the model's capacity to resolve fine chemical variations, providing a practical tool to guide experimental design and contribute to the development of BioFETs. We further identify notable areas for future work where we will engage with more rigorous modelling, utilizing our model outputs to guide experimental design and further validate our device modelling framework.