Machine Learning-Aided Molecular Dynamics Simulation for Prediction of Binding Kinetics

Abstract

Chemical sensors provide new solutions to address some of the world’s biggest challenges, including climate change, energy and healthcare. Understanding molecule binding kinetics and thermodynamics is essential in enhancing the design and functionality of chemical sensors. To contribute to this field, we have developed a numerical framework to predict the binding kinetics without requiring experimental inputs. Once the target molecules and fictional surface are identified, the details alongside the environment and mass transport are included as input to this code. The output would be the predictive model of target molecule behaviour passing by the surface. This framework comprises an all-atom molecular dynamics model and a Bayesian machine learning model for predicting affinity. Different predictive models have been trained, and the Bayesian-based Gaussian process regression (GPR) best predicts the binding reaction amongst them all. The predictive model is validated for an aluminium-based platform. The proposed numerical framework has the potential to be generalised and, therefore, contribute to future low-cost binding reaction estimations, providing a valuable tool for industry and experimentalists.