Approaches for computing small molecule binding free energies based on molecular simulations are now regularly being employed by academic and industry practitioners to study receptor-ligand systems and prioritize the synthesis of small molecules for ligand design. Given the variety of methods and implementations available, it is natural to ask how the convergence rates and final predictions of these methods compare. In this study, we describe the concept and results for the SAMPL6 SAMPLing challenge, the first challenge from the SAMPL series focusing on the assessment of convergence properties and reproducibility of binding free energy methodologies. We provided parameter files, partial charges, and multiple initial geometries for two octa-acid (OA) and one cucurbit[8]uril (CB8) host-guest systems. Participants submitted binding free energy predictions as a function of the number of force and energy evaluations for seven different alchemical and physical-pathway (i.e., potential of mean force and weighted ensemble of trajectories) methodologies implemented with the GROMACS, AMBER, NAMD, or OpenMM simulation engines. To rank the methods, we developed an efficiency statistic based on bias and variance of the free energy estimates. For the two small OA binders, the free energy estimates computed with alchemical and potential of mean force approaches show relatively similar variance and bias as a function of the number of energy/force evaluations, with the attach-pull-release (APR), GROMACS expanded ensemble, and NAMD double decoupling submissions obtaining the greatest efficiency. The differences between the methods increase when analyzing the CB8-quinine system, where both the guest size and correlation times for system dynamics are greater. For this system, nonequilibrium switching (GROMACS/NS-DS/SB) obtained the overall highest efficiency. Surprisingly, the results suggest that specifying force field parameters and partial charges is insufficient to generally ensure reproducibility, and we observe differences between seemingly converged predictions ranging approximately from 0.3 to 1.0 kcal/mol, even with almost identical simulations parameters and system setup (e.g., Lennard-Jones cutoff, ionic composition). Further work will be required to completely identify the exact source of these discrepancies. Among the conclusions emerging from the data, we found that Hamiltonian replica exchange-while displaying very small variance-can be affected by a slowly-decaying bias that depends on the initial population of the replicas, that bidirectional estimators are significantly more efficient than unidirectional estimators for nonequilibrium free energy calculations for systems considered, and that the Berendsen barostat introduces non-negligible artifacts in expanded ensemble simulations.

As most relevant motions in biomolecular systems are inaccessible to conventional molecular dynamics simulations, algorithms that enhance sampling of rare events are indispensable. Increasing interest in intrinsically disordered systems and the desire to target ensembles of protein conformations (rather than single structures) in drug development motivate the need for enhanced sampling algorithms that are not limited to “two-basin” problems, and can efficiently determine structural ensembles. For systems that are not well-studied, this must often be done with little or no information about the dynamics of interest. Here we present a novel strategy to determine structural ensembles that uses dynamically defined sampling regions that are organized in a hierarchical framework. It is based on the weighted ensemble algorithm, where an ensemble of copies of the system (“replicas”) is directed to new regions of configuration space through merging and cloning operations. The sampling hierarchy allows for a large number of regions to be defined, while using only a small number of replicas that can be balanced over multiple length scales. We demonstrate this algorithm on two model systems that are analytically solvable and examine the 10-residue peptide chignolin in explicit solvent. The latter system is analyzed using a configuration space network and novel hydrogen bonds are found that facilitate folding.

Ligand (un)binding kinetics is being recognized as a determinant of drug specificity and efficacy in an increasing number of systems. However, the calculation of kinetics and the simulation of drug unbinding is more difficult than computing thermodynamic quantities, such as binding free energies. Here we present the first full simulations of an unbinding process at pharmacologically relevant time scales (11 min), without the use of biasing forces, detailed prior knowledge, or specialized processors using the weighted ensemble based algorithm, WExplore. These simulations show the inhibitor TPPU unbinding from its enzyme target soluble epoxide hydrolase, which is a clinically relevant target that has attracted interest in kinetics optimization in order to increase efficacy. We make use of conformation space networks that allow us to conceptualize unbinding not just as a linear process, but as a network of interconnected states that connect the bound and unbound states. This allows us to visualize patterns in hydrogen-bonding, solvation, and nonequilibrium free energies, without projection onto progress coordinates. The topology and layout of the network reveal multiple unbinding pathways, and other rare events, such as the reversal of ligand orientation within the binding site. Furthermore, we make a prediction of the transition state ensemble, using transition path theory, and identify protein-ligand interactions which are stabilizing to the transition state. Additionally, we uncover trends in ligand and binding site solvation that corroborate experimental evidence from more classical structure kinetics relationships and generate new questions as to the role of drug modifications in kinetics optimization. Finally, from only 6 μs of simulation time we observed 75 unbinding events from which we calculate a residence time of 42 s, and a standard error range of 23 to 280 s. This nearly encompasses the experimental residence time 11 min (660 s). In addition to the insights to sEH inhibitor unbinding, this study shows that simulations of complex processes on timescales as long as minutes are becoming feasible for more researchers to perform.

We report simulations of full ligand exit pathways for the trypsin-benzamidine system, generated using the sampling technique WExplore. WExplore is able to observe millisecond-scale unbinding events using many nanosecond-scale trajectories that are run without introducing biasing forces. The algorithm generates rare events by dividing the coordinate space into regions, on-the-fly, and balancing computational effort between regions through cloning and merging steps, as in the weighted ensemble method. The averaged exit flux yields a ligand exit rate of 180 μs, which is within an order of magnitude of the experimental value. We obtain broad sampling of ligand exit pathways, and visualize our findings using conformation space networks. The analysis shows three distinct exit channels, two of which are formed through large, rare motions of the loop regions in trypsin. This broad set of ligand-bound poses is then used to investigate general properties of ligand binding: we observe both a direct stabilizing effect of ligand-protein interactions and an indirect destabilizing effect on intraprotein interactions that is induced by the ligand. Significantly, the crystallographic binding poses are distinguished not only because their ligands induce large stabilizing effects, but also because they induce relatively low indirect destabilizations.