Resolving Protein Conformational Plasticity and Substrate Binding Through the Lens of Machine-Learning

A long-standing target in elucidating the biomolecular recognition process is the identification of binding-competent conformations of the receptor protein. However, protein conformational plasticity and the stochastic nature of the recognition processes often preclude the assignment of a specific protein conformation to an individual ligand-bound pose. In particular, we consider multi-microsecond long Molecular dynamics simulation trajectories of ligand recognition process in solvent-inaccessible cavity of two archtypal systems: L99A mutant of T4 Lysozyme and Cytochrome P450. We first show that if the substrate-recognition occurs via long-lived intermediate, the protein conformations can be automatically classified into substrate-bound and unbound state through an unsupervised dimensionality reduction technique. On the contrary, if the recognition process is mediated by selection of transient protein conformation by the ligand, a clear correspondence between protein conformation and binding-competent macrostates can only be established via a combination of supervised machine learning (ML) and unsupervised dimension reduction approach. In such scenario, we demonstrate that an a priori random forest based supervised classification of the simulated trajectories recognition process would help characterize key amino-acid residue-pairs of the protein that are deemed sensitive for ligand binding. A subsequent unsupervised dimensional reduction via time-lagged independent component analysis of the selected residue-pairs would delineate a conformational landscape of protein which is able to demarcate ligand-bound pose from the unbound ones. As a key breakthrough, the ML-based protocol would identify distal protein locations which would be allosterically important for ligand binding and characterise their roles in recognition pathways.