514
S. von Grafenstein et al.
kinase 3 and β-secretase [76]. This might be one of the reasons why FlexE was
more frequently applied on ensembles of alternative X-ray structures and not on
MD-derived ensembles where larger transitions are expected.
Likewise, ensemble-based grid representation of binding sites is rarely used for
MD-derived ensembles. Ensemble-based grid representations, also called composite-grids, are one of the oldest concepts to include protein flexibility within docking
[74]. Technically, the derivation of one grid from an MD ensemble follows the same
approach as for multiple X-ray structures or NMR ensembles. Different averaging
strategies were evaluated to calculate interaction potentials of the grid points for
experimental ensembles [74, 77]. Broughton averaged the grid representation of
conformations extracted from short MD simulations (< 100 ps) of cyclooxygenase
2 and observed an improvement over docking to a rigid protein in terms of ranking
using the docking program FLOG [78].
Obviously, merging strategies are especially time efficient [79]. A single docking
run covers all information from an ensemble, so the time per docked compound is
minimized compared to methods explicitly relying on multiple independent receptor conformations. However, taking into account the computational need for the
generation of an ensemble using all-atom MD simulations, the argument of computational demand while docking loses weight. Additionally, atomistic information of
specific conformational states sampled during MD simulation is lost by the unification into one representation. With these aspects in mind it is not surprising that alternative strategies of ensemble representation are preferentially used in combination
with MD-generated ensembles.
Parallel docking to multiple conformations selected from the trajectory is frequently used in combination with MD simulations. The periodic extraction
of snapshots from an MD simulation is a straightforward way to select multiple
conformations. Subsequently, molecular docking is performed on the individual
snapshots independently from each other. The continuous and unbiased selection
should ideally represent the state space sampled in the simulation. In consequence,
the resulting collection of snapshots represents a valid thermodynamic ensemble
and the subsequent ensemble of docking results are interpreted via statistical thermodynamics (see Sect. 15.2.2.3).
A protocol using time-dependent extraction of conformers from MD simulations
and subsequent docking is developed by the McCammon group and known as “relaxed-complex-scheme” (RCS) [41, 80, 81, 82]. Classically, RCS applications use
AutoDock, an established docking program for flexible ligand docking based on a
genetic algorithm. Recently, AutoDock Vina a new implementation of the algorithm
is used as it is more suitable for parallel application [83]. A major success of this
protocol was the identification of a cavity next to the binding site in HIV integrase
[81] allowing the development of inhibitors specifically addressing this cavity. The
approved drug raltegravir was developed based on the binding mode suggested by
Schames et al. Besides this one significant contribution to drug discovery, several
more applications of the RCS are described in literature and represent MD-derived
ensemble docking studies using periodic snapshot extraction.
S. von Grafenstein et al.
kinase 3 and β-secretase [76]. This might be one of the reasons why FlexE was
more frequently applied on ensembles of alternative X-ray structures and not on
MD-derived ensembles where larger transitions are expected.
Likewise, ensemble-based grid representation of binding sites is rarely used for
MD-derived ensembles. Ensemble-based grid representations, also called composite-grids, are one of the oldest concepts to include protein flexibility within docking
[74]. Technically, the derivation of one grid from an MD ensemble follows the same
approach as for multiple X-ray structures or NMR ensembles. Different averaging
strategies were evaluated to calculate interaction potentials of the grid points for
experimental ensembles [74, 77]. Broughton averaged the grid representation of
conformations extracted from short MD simulations (< 100 ps) of cyclooxygenase
2 and observed an improvement over docking to a rigid protein in terms of ranking
using the docking program FLOG [78].
Obviously, merging strategies are especially time efficient [79]. A single docking
run covers all information from an ensemble, so the time per docked compound is
minimized compared to methods explicitly relying on multiple independent receptor conformations. However, taking into account the computational need for the
generation of an ensemble using all-atom MD simulations, the argument of computational demand while docking loses weight. Additionally, atomistic information of
specific conformational states sampled during MD simulation is lost by the unification into one representation. With these aspects in mind it is not surprising that alternative strategies of ensemble representation are preferentially used in combination
with MD-generated ensembles.
Parallel docking to multiple conformations selected from the trajectory is frequently used in combination with MD simulations. The periodic extraction
of snapshots from an MD simulation is a straightforward way to select multiple
conformations. Subsequently, molecular docking is performed on the individual
snapshots independently from each other. The continuous and unbiased selection
should ideally represent the state space sampled in the simulation. In consequence,
the resulting collection of snapshots represents a valid thermodynamic ensemble
and the subsequent ensemble of docking results are interpreted via statistical thermodynamics (see Sect. 15.2.2.3).
A protocol using time-dependent extraction of conformers from MD simulations
and subsequent docking is developed by the McCammon group and known as “relaxed-complex-scheme” (RCS) [41, 80, 81, 82]. Classically, RCS applications use
AutoDock, an established docking program for flexible ligand docking based on a
genetic algorithm. Recently, AutoDock Vina a new implementation of the algorithm
is used as it is more suitable for parallel application [83]. A major success of this
protocol was the identification of a cavity next to the binding site in HIV integrase
[81] allowing the development of inhibitors specifically addressing this cavity. The
approved drug raltegravir was developed based on the binding mode suggested by
Schames et al. Besides this one significant contribution to drug discovery, several
more applications of the RCS are described in literature and represent MD-derived
ensemble docking studies using periodic snapshot extraction.