Alchemical free energy methods represent a powerful tool in computer-aided drug design and can computationally evaluate proposed modifications to a lead compound prior to their testing in the lab. Traditional methods, including free energy perturbation theory or thermodynamic integration (FEP/TI), however, are inherently limited to the examination of only one perturbation at a time. This scalability limitation is surmounted with λ-dynamics, which can examine multiple alchemical modifications simultaneously within a single molecular dynamics simulation. Using λ-dynamics, it has been shown that tens to hundreds of alchemical states of a lead compound can be explored with modest resource consumption. This work introduces a new variant of λ-dynamics, called discrete-Gibbs sampler-based λ-dynamics (d-GSλD). This new approach allows many substituents to be explored at multiple sites around the common core of a lead compound and can use λ dependent potentials, such as soft-core Lennard Jones potentials, to compute relative free energy differences. Energy calculations at all discrete λ states are performed on-the-fly, allowing real-time convergence metrics to be determined as the simulation is run. Accurate free energy differences are computed using FastMBAR, a novel formulation of the widely used multistate Bennett acceptance ratio (MBAR) free energy estimator. Overall, d-GSλD combines the simplicity of setup associated with FEP/TI methods, the accuracy of MBAR, and the scalability of λ-dynamics to yield a rigorous and efficient free energy method capable of rapidly exploring large combinatorial chemical spaces. These characteristics will be demonstrated for a variety of small molecule and protein-ligand binding examples as it relates to drug design.

Exploring combinatorial chemical spaces with a discrete-Gibbs sampler-based λ-dynamics approach

2019 Fall National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.