Improved Ligand Binding Energies Derived from Molecular Dynamics: Replicate Sampling Enhances the Search of Conformational Space

Abstract

Does a single molecular trajectory provide an adequate sample conformational space? Our calculations indicate that for Molecular Mechanics--Poisson-Boltzmann Surface Area (MM-PBSA) measurement of protein ligand binding, a single molecular dynamics trajectory does not provide a representative sampling of phase space. For a single trajectory, the binding energy obtained by averaging over a number of molecular dynamics frames in an equilibrated system will converge after an adequate simulation time. A separate trajectory with nearly identical starting coordinates (1% randomly perturbed by 0.001 Å), however, can lead to a significantly different calculated binding energy. Thus, even though the calculated energy converges for a single molecular dynamics run, the variation across separate runs implies that a single run inadequately samples the system. The divergence in the trajectories is reflected in the individual energy components, such as the van der Waals and the electrostatics terms. These results indicate that the trajectories sample different conformations that are not in rapid exchange. Extending the length of the dynamics simulation does not resolve the energy differences observed between different trajectories. By averaging over multiple simulations, each with a nearly equivalent starting structure, we find the standard deviation in the calculated binding energy to be ∼1.3 kcal/mol. The work presented here indicates that combining MM-PBSA with multiple samples of the initial starting coordinates will produce more precise and accurate estimates of protein/ligand affinity.

Unpicking the Cause of Stereoselectivity in Actinorhodin Ketoreductase Variants with Atomistic Simulations

“…(The 24 ps-interval at which snapshots are extracted is roughly in line with previous MM/PBSA and MM/GBSA indications, but could be reduced). [83][84][85] To study FIII, we employed a semiempirical QM/MM approach, to avoid specific parameterization of models (e.g. as required for empirical valence bond methods) 36 and limit computational cost (e.g.…”

Including Explicit Water Molecules as Part of the Protein Structure in MM/PBSA Calculations

“…We ran replicate simulations, each with slightly perturbed initial coordinates, 13 Table S2B, Supporting Information) whereas some ligands were not in the first two runs (Run 1 and 2) because, originally, the scope of the investigation was limited to selected ligands and was expanded to all ligands later.…”

Can MM/GBSA calculations be sped up by system truncation?

“…To obtain statistically converged results, we used 40 independent simulations for each of the ligands, by assigning different starting velocities to atoms (i.e., 48 ns simulations and 1600 energy calculations in total for each ligand) . Previous studies have shown that many short independent simulations cover the phase space more effective than a single long simulation and give a more representative estimate of the statistical uncertainty . Reported uncertainties are the standard deviation over the results (averages over the 40 snapshots) from the 40 independent simulations, divided by .…”

“…[50] Previous studies have shown that many short independent simulations cover the phase space more effective than a single long simulation and give a more representative estimate of the statistical uncertainty. [35,[50][51][52][53][54][55] Reported uncertainties are the standard deviation over the results (averages over the 40 snapshots) from the 40 independent simulations, divided by ffiffiffiffiffi 40 p .…”