529
15 (How to) Profit from Molecular Dynamics-based Ensemble Docking
repeats. As a criterion for the stability they calculate the residence time of the ligand. They showed that on average docking poses close to the native structure have
a higher residence time and non-native poses can be eliminated based on a low
residence time. Consequently, they suggest using the residence time as a criterion
for scoring of ligand poses [126].
Molecular dynamics allow estimating binding free energies of small molecules
to proteins in a physics-based framework. These methods predict Gibbs free energies of protein–ligand binding. In post-processing they can complement the estimates from docking and be used for rescoring. Depending on the level of accuracy necessary, several different methods for free energy calculation can be applied
[127]. Free energy calculations are especially suited to provide relative binding free
energies of congeneric ligand series as errors introduced, e.g., with the force field
are canceling. So, some of these techniques are applied in lead optimization where
series of similar compounds are evaluated (thermodynamic integration, free energy
perturbation). In virtual screening diverse chemical entities need to be evaluated.
This task can be performed with MM/PBSA, MM/GBSA and LIE approach.
Post-processing of molecular dynamics trajectories allows to perform molecular
mechanics/Poisson-Boltzmann Surface Area (MM/PBSA) [128] or faster and more
approximate molecular mechanics/Generalized Born Surface Area (MM/GBSA)
[129] calculations. These statistical methods estimate binding energies by summing
up in vacuo force field energies, polar and non-polar solvation energies as well as
entropic contributions from normal mode analysis. The difference between both
approaches lies in the calculation of the polar solvation energy, with GBSA using
more approximations to yield a numerical estimate of the electrostatic solvation
energy than PBSA. The non-polar solvation term captures the hydrophobic effect,
forcing apolar groups to bury inside of a protein. A final free energy estimate is generated by subtracting respective energies of free receptor and free ligand from the
complex. MM/PBSA was used in first applications of the relaxed-complex-scheme
to rescore the best poses for each ligand. Artificially high binding energies predicted
by the force field-based scoring function of AutoDock could be avoided [80]. Negri
et al. applied docking and subsequent MD simulations evaluated by MM/PBSA to
investigate the binding mode of bis(hydroxyl)phenylarenes to 17β-hydroxysteroid
dehydrogenase type 1 [130]. On a larger scale Okimoto et al. suggested to use docking and subsequent scoring with MM/PBSA in virtual screening application. The
authors showed that MM/PBSA-based scoring enables to identify the near-native
pose for a trypsin inhibitor whereas docking scores failed to rank this pose accurately. ROC values improved for 3 out of 4 test cases using MM/PBSA over a docking
protocol using the program GOLD [124]. Performing MD-based free energy estimation for multiple ligands and even for multiple poses per ligand (as done in this
study) leads to an explosion of computational cost. The application of this approach
in high-throughput scale is only feasible for facilities where tailored computational
infrastructure is accessible [124].
An alternative approach is the linear interaction energy (LIE) [131]. Herein, the
binding energy is estimated via summation of weighted contributions of ensembleaveraged electrostatics and van-der-Waals interactions. The difference between
15 (How to) Profit from Molecular Dynamics-based Ensemble Docking
repeats. As a criterion for the stability they calculate the residence time of the ligand. They showed that on average docking poses close to the native structure have
a higher residence time and non-native poses can be eliminated based on a low
residence time. Consequently, they suggest using the residence time as a criterion
for scoring of ligand poses [126].
Molecular dynamics allow estimating binding free energies of small molecules
to proteins in a physics-based framework. These methods predict Gibbs free energies of protein–ligand binding. In post-processing they can complement the estimates from docking and be used for rescoring. Depending on the level of accuracy necessary, several different methods for free energy calculation can be applied
[127]. Free energy calculations are especially suited to provide relative binding free
energies of congeneric ligand series as errors introduced, e.g., with the force field
are canceling. So, some of these techniques are applied in lead optimization where
series of similar compounds are evaluated (thermodynamic integration, free energy
perturbation). In virtual screening diverse chemical entities need to be evaluated.
This task can be performed with MM/PBSA, MM/GBSA and LIE approach.
Post-processing of molecular dynamics trajectories allows to perform molecular
mechanics/Poisson-Boltzmann Surface Area (MM/PBSA) [128] or faster and more
approximate molecular mechanics/Generalized Born Surface Area (MM/GBSA)
[129] calculations. These statistical methods estimate binding energies by summing
up in vacuo force field energies, polar and non-polar solvation energies as well as
entropic contributions from normal mode analysis. The difference between both
approaches lies in the calculation of the polar solvation energy, with GBSA using
more approximations to yield a numerical estimate of the electrostatic solvation
energy than PBSA. The non-polar solvation term captures the hydrophobic effect,
forcing apolar groups to bury inside of a protein. A final free energy estimate is generated by subtracting respective energies of free receptor and free ligand from the
complex. MM/PBSA was used in first applications of the relaxed-complex-scheme
to rescore the best poses for each ligand. Artificially high binding energies predicted
by the force field-based scoring function of AutoDock could be avoided [80]. Negri
et al. applied docking and subsequent MD simulations evaluated by MM/PBSA to
investigate the binding mode of bis(hydroxyl)phenylarenes to 17β-hydroxysteroid
dehydrogenase type 1 [130]. On a larger scale Okimoto et al. suggested to use docking and subsequent scoring with MM/PBSA in virtual screening application. The
authors showed that MM/PBSA-based scoring enables to identify the near-native
pose for a trypsin inhibitor whereas docking scores failed to rank this pose accurately. ROC values improved for 3 out of 4 test cases using MM/PBSA over a docking
protocol using the program GOLD [124]. Performing MD-based free energy estimation for multiple ligands and even for multiple poses per ligand (as done in this
study) leads to an explosion of computational cost. The application of this approach
in high-throughput scale is only feasible for facilities where tailored computational
infrastructure is accessible [124].
An alternative approach is the linear interaction energy (LIE) [131]. Herein, the
binding energy is estimated via summation of weighted contributions of ensembleaveraged electrostatics and van-der-Waals interactions. The difference between