Scoring ligand/protein interactions using a novel high performance implementation of DFT quantum mechanics

Project Summary In the past several years, dramatic advances in available compute resources have finally enabled computational chemistry to significantly impact hit-to-lead discovery. Most significantly, free energy perturbation (FEP) calculations have demonstrated high value for rank-ordering the binding of multiple ligands to a common protein receptor, thereby allowing acceptably accurate, cost-effective predictions with reasonable turnaround time. Use of this type of approach has been credited for speeding the hit-to-lead process by a factor of more. (E.g. the computationally-driven Nimbus / Schrodinger collaboration that yielded an Acetyl-CoA carboxylase inhibitor clinical candidate in 18 months and that was sold to Gilead in a deal worth as much as $1.2 billion). But, while approaches like FEP are demonstrably useful, they are limited in application to certain types of systems. This is because they rely on the use of a classical energy function, which attempts to mimic quantum effects with an approximate form that needs to be parameterized. Ideally, one could, instead, simply use quantum mechanics (QM), and avoid the significant limitations (in domain of applicability and accuracy) that arise from using a fitted force field. However, until very recently, high accuracy quantum mechanics calculations could not be performed on full ligand/protein systems. We have now demonstrated, for the first time, how a unique distributed memory implementation of high accuracy quantum mechanics (density functional theory (DFT)) allows ligand/protein binding to be evaluated entirely in the quantum domain. These calculations can be carried out on commodity hardware (Cloud computers), can finish in less than an hour and at a cost of less than roughly $10. Having removed the limitations of the classical force field, these quantum calculations promise to be more broadly applicable, faster, and cheaper than the FEP equivalents. In this grant, we describe a set of efforts that are required to more fully characterize and validate this approach, and to move from the potential energies that are directly determined in quantum mechanics to free energy estimates that are more directly comparable to experimental measurements. We will utilize a benchmark set of consistently-measured experimental binding data for ligand/protein interactions for a variety of proteins to test and improve our DFT method. Part of our work will focus on identifying the optimal approach for very large protein systems. While the entire ligand/protein complex can be treated quantum mechanically with our approach for proteins of size andlt;~ 3000 atoms, for larger systems, we will need to identify a hybrid approach that treats a large radius around the binding site using QM, with the remainder of the system treated using a computationally less expensive approach. We also will look to identify corrections to the QM calculated energies that form a bridge between the fully QM potential energy and free energy that reflects entropic effects. Particular focus will be on a method such as 3D-2FT, which uses an atomistic statistical mechanical approach to estimate the entropic cost of displacing waters in the protein binding site. If successful, these new quantum calculations have the potential to further broaden the importance and value of computational chemistry in the drug discovery pipeline.Project Narrative Drug discovery is an expensive and risky process that traditionally takes more than a decade and 2.6 billion dollars, between the time when a target is identified until when a drug appears on the market—and many projects fail to make it to market at all. Computational chemistry has fundamentally improved both the time and cost of small molecule drug discovery, but the best methods in current wide use still fall short in the sense that they are not broadly and reliably enough applicable to systems of interest. We propose a new approach based on quantum mechanics (QM/DFT) that promises to improve how widely accurate computational methods can be applied to drug discovery, and further reduce the time and cost to bring a pharmaceutical small molecule drug to market.