Development and Application of Grand Canonical Nonequilibrium Candidate Monte Carlo for in silico Prediction of Fragment Binding Sites, Modes, and Affinities

Abstract

Structure and fragment-based drug design are increasingly popular approaches to drug discovery. Computational tools have become integral to these campaigns and provide a route to library design, virtual screening, property prediction, identifying putative binding sites, elucidating binding geometries, and predicting accurate binding affinities. This thesis discusses various molecular simulation methods and assesses their applicability to these drug discovery regimes.

Molecular dynamics-based simulations are a useful tool in computer-aided drug design but are often limited by sampling issues related to the simulation timescales obtainable. Here, we develop, implement, validate, and test the application of grand canonical nonequilibrium candidate Monte Carlo (GCNCMC) to accurately predict the binding sites, modes, and affinities of fragment-like molecules. To this end, we develop the Python module, grandlig. GCNCMC simulations can accurately predict the location of small molecules in protein-ligand systems by attempting the insertion and deletion of molecules to, or from, a region of interest; each proposed move is subject to a rigorous acceptance test based on the thermodynamic properties of the system.

We first set the scene and highlight the limitations of basic MD simulations by applying a variety of methods to the ERK2 protein. The theory and development of ligand-based GCNCMC is then presented with a rigorous validation of the method. The subsequent chapters then present various ways in which GCNCMC can be used to enhance the drug discovery pipeline by applying the method to two protein-ligand systems, T4L99A and MUP1. We demonstrate the ability of fragment-based GCNCMC to rapidly and reliably find experimental fragment binding sites, show that the method can accurately sample multiple fragment binding modes without any prior knowledge of their existence, and finally demonstrate the method's ability as a free energy estimator.

We present two novel applications of GCNCMC; the integration of GCNCMC into mixed solvent MD, a popular method for binding site identification, and as a fragment screening tool. In both cases, we observe promising results and outline steps for the future which could make this method a powerful tool in the computational-aided drug design arsenal.

More information

Identifiers

ORCID for Jonathan Essex: orcid.org/0000-0003-2639-2746

Catalogue record

Export record