Quantum mechanical bespoke force fields for computer-aided drug design

Abstract

The ability to accurately model complex biological processes such as protein-ligand binding with an atomistic level of detail is critical to their thorough understanding. Typically a molecular mechanics simulation is used, which represents the system using a force eld that is a physically motivated linear combination of empirically parameterised potentials. Traditionally their parameterisation has involved the recreation of experimental and quantum mechanical data for a target set of representative structures, ranging from small molecules to peptides. This potentially limits the progress of general transferable force elds to time and labour-intensive incremental improvements. In this thesis, we aim to challenge this \parameterise once and transfer" philosophy, with that of a transferable parametrisation methodology that can be readily applied to new systems with a consistent level of accuracy. We collect together recently developed force eld parameterisation techniques from the literature to develop a protocol suitable to derive virtually all required force eld parameters for small molecules directly from quantum mechanics. This protocol forms the basis of the QUantum mechanical BEspoke force eld (QUBE) and is delivered to users through a reliable and extensible software toolkit named QUBEKit. Here we extensively benchmark the methodology and software presented through typical force eld performance metrics which involve the prediction of thermodynamical properties of small organic molecules. In this regard, we achieve very competitive accuracy with popular general transferable force elds such as OPLS which have been extensively optimised to reproduce such properties. We also demonstrate how the QUBE force eld is a suitable alternative in a computer-aided drug design setting via the retrospective calculation of the relative binding free energies of 17 inhibitors of p38 MAP kinase. Again good agreement with both experiment and transferable force elds is achieved despite this being the rst generation of the force eld. The results of this work are then particularly important to those studying systems which are not covered or inaccurately represented by standard transferable force elds, as we present an accurate framework towards their complete parameterisation.