UC Irvine

Free energy perturbation methods serve an important role in drug discovery by providing accurate predictions of binding affinity, solubility, and other quantities.

However, in order for the free energy estimates to be accurate, the system must be able to sample all the relevant low energy states during the course of a simulation.

This proves to be challenging for binding affinity calculations in particular, since there can be many different potential binding modes, and binding modes are slow to interconvert at simulation timescales.

It is possible to treat each binding mode separately, perform a free energy calculation on each binding mode, and then combine the results into a total free energy prediction, but the computational cost of free energy calculations does make this parallelization approach feasible.

Another alternative approach would be to use a method that could sample between the different binding modes efficiently and produce accurate estimates of the populations of each of those binding modes.

This information, along with a binding free energy calculation on one of the binding modes, would allow the estimation of the overall binding free energy.

In order to sample between binding modes efficiently, I helped develop a new method that uses nonequilibrium candidate Monte Carlo (NCMC) to remove the ligand and reinsert it in a new binding site with a center of mass rotation to further improve sampling.

I validated this methodology on T4 lysozyme L99A, a model protein for binding, and was able to show that it enhanced the sampling of the binding modes of toluene and 3-iodotolune.

Following these results, I helped create the BLUES (Binding modes of Ligands Using Enhanced Sampling) software package to facilitate the use of this technique.

Originally BLUES only could further improve binding mode sampling with a center of mass rotation Monte Carlo (MC) move, which limited it's applicability to small, rigid ligands.

To further improve binding mode sampling, I also developed a new type of MC move called molecular Darting (MolDarting) to sample specific binding modes.

Through MolDarting it is possible to sample predefined conformations–obtained by docking, for example–and reversibly sample them in a MC framework.

MolDarting also opens up the ability to even sample ligand binding modes in separate binding sites.

We validated this move on an alanine-valine dipeptide system, as well the previouly explored T4 lysozyme and attempted to sample all the potential binding sites in HIV integrase.