Modeling sensorgrams issued by a surface plasmon resonance biosensor.

Abstract

An extensive computational framework was developed and validated to model and address limitations in the analysis of Surface Plasmon Resonance (SPR) sensorgrams. The framework contemplates a model for bulk refractive index changes, a spatially-resolved model for surface heterogeneity, and advanced parameter estimation algorithms, e.g., Recursive Least Squares (RLS) for real-time analysis and a novel LSTM-based Physics-Informed Neural Network (PINN). Validation was performed using benchmark literature data and new experimental data from a diffractive optical coupling (DOCE)-based SPR sensor. The results show that the framework accurately simulates the bulk responses for NaOH and HCl. The RLS algorithm achieved fits with an average iteration of 6.1 x 10−5 s but yielded effective, non-physical kinetic constants. The PINN, however, identified kinetic parameters from noisy data with 4-9% relative error. The heterogeneous model successfully recovered "kinetic maps" from noisy data. Additionally, a statistical analysis provided quantitative proof that sensor calibration and material quality critically impact data reproducibility. This work delivers a validated suite of tools for SPR analysis, from efficient real-time prediction to physically rigorous characterization of heterogeneous surface interactions using AI-driven methods.