Solving the sampling problem in molecular simulations by Sequential Monte Carlo

Molecular simulations are an essential tool in the design of new drugs and materials, and are widely used to provide atomistic detail to augment low-resolution experimental data. In a molecular simulation, the atoms in the system move in response to the energy and forces acting on them, and by examining the molecular arrangements adopted, new optimised molecules and materials are designed. For example, by exploring possible geometries of a drug binding to its receptor, new interactions may be identified and exploited, leading to better drugs with higher affinity or better selectivity.

The extent to which current simulations are able to explore these new binding geometries is very limited. Conventional molecular dynamics is very efficient at sampling a particular binding geometry, but the large kinetic barriers separating other possible binding geometries mean that these are seldom observed in the simulations, if at all - the simulation is myopic and trapped. Brute force - getting a bigger computer - is a solution for some, but this represents a massive financial investment that is beyond the capability of the overwhelming majority of workers. We therefore need to be smarter. There are a range of enhanced sampling algorithms, which seek to solve this problem of poor sampling. They typically work in one of two ways. They either reduce the energy barrier between the possible stable binding geometries, so that the simulation can smoothly move between them, or they add energy to the simulation, so that the barriers may be crossed naturally. However, all these methods suffer from disadvantages that make them inefficient and requiring considerable system-specific optimisation.

This proposal seeks to solve the sampling problem, by developing and applying a widely used sampling procedure from statistics - Sequential Monte Carlo. This approach will be general and adaptable. In doing so this high-risk and adventurous project will deliver robust new molecular simulation methodology to transform the discovery of new molecules and materials.

Molecular-SMC is adaptive, efficient, and free of the need to know a priori the detailed structural rearrangements of protein-ligand systems. Here the method will be developed and applied to two pressing problems in drug discovery - in protein-ligand docking, where the particular problem of variable hydration will be addressed, and in the more rigorous area of binding free energy calculations, where subtle modifications to the ligand can bring about substantial changes in binding geometry.