520
S. von Grafenstein et al.
the use of collective variables sampling the plasticity of the protein in presence of
a ligand to allow ligand-specific adaption of the protein conformation [63]. Selection of the ligand-specific conformations from a set of pre-generated conformers
corresponds to the challenge of cross docking into alternative X-ray structures [94].
Applications focusing on MD structures used a consensus representation for all
conformations as starting point and then included the sampling of individual conformations within the simplex optimization of the docking pose using the program
DOCK [95]. In a review, Totrov and Abagyan suggest that other optimization algorithms might be more suitable to address the optimization of this parameter [8].
The concept of four dimensional (4D) docking follows a similar concept using an
optimization based on MC sampling established by Bottegoni et al. for the docking
program ICM [96]. For a large set of X-ray derived ensembles, they showed a similar performance in reproducing the native binding pose in comparison to the parallel
use of the X-ray ensembles without linear increase of computational time with number of structures. However, the method was neither evaluated for the performance
to discriminate ligands from decoys nor was it applied on MD-derived ensembles.
Using structures generated with a normal mode approach Leis et al. included the
selection of the structure in the genetic algorithm implemented in AutoDock [66].
Studying binding to protein kinase A, they showed that this approach worked best
to reproduce the binding mode for cases where the additional degrees of freedom,
e.g., flexible torsion angles in the ligand are limited. The presented cases did not
use conformations from an MD simulation. However, apart from sampling issues no
limitations for the application on MD-derived ensembles are known by now.
The ongoing research in the field indicates that the optimal way to tackle the selection of conformers is not yet identified within the variety of strategies followed.
Moreover, this issue is not an isolated challenge within the workflow of ensemble
docking, but is directly related to the evaluation of docking results. Depending on
the strategy followed for structure selection docking results will either be a single
hit list or a collection of several hit lists. Subsequent pooling strategies for multiple
hit lists depend on the principal approach employed for selection of receptor representation.
15.2.2 Molecular Docking
15.2.2.1 Docking Algorithms and Scoring Functions
Molecular docking can be considered as one of the major pillars of structure-based
drug design. Docking flexible ligands, mostly small organic molecules, to a specific site of a receptor representing a rigid bio-molecular structure is an established
procedure [97, 98, 99, 100, 101]. Therefore, there is a critical awareness for the
potential as well as the limitations of this technique [92, 102]. Well accepted but still
open challenges are the inclusion of backbone flexibility (which is subject of this
overview), treatment of solvent and solvent-related effects [49], and quantitative
S. von Grafenstein et al.
the use of collective variables sampling the plasticity of the protein in presence of
a ligand to allow ligand-specific adaption of the protein conformation [63]. Selection of the ligand-specific conformations from a set of pre-generated conformers
corresponds to the challenge of cross docking into alternative X-ray structures [94].
Applications focusing on MD structures used a consensus representation for all
conformations as starting point and then included the sampling of individual conformations within the simplex optimization of the docking pose using the program
DOCK [95]. In a review, Totrov and Abagyan suggest that other optimization algorithms might be more suitable to address the optimization of this parameter [8].
The concept of four dimensional (4D) docking follows a similar concept using an
optimization based on MC sampling established by Bottegoni et al. for the docking
program ICM [96]. For a large set of X-ray derived ensembles, they showed a similar performance in reproducing the native binding pose in comparison to the parallel
use of the X-ray ensembles without linear increase of computational time with number of structures. However, the method was neither evaluated for the performance
to discriminate ligands from decoys nor was it applied on MD-derived ensembles.
Using structures generated with a normal mode approach Leis et al. included the
selection of the structure in the genetic algorithm implemented in AutoDock [66].
Studying binding to protein kinase A, they showed that this approach worked best
to reproduce the binding mode for cases where the additional degrees of freedom,
e.g., flexible torsion angles in the ligand are limited. The presented cases did not
use conformations from an MD simulation. However, apart from sampling issues no
limitations for the application on MD-derived ensembles are known by now.
The ongoing research in the field indicates that the optimal way to tackle the selection of conformers is not yet identified within the variety of strategies followed.
Moreover, this issue is not an isolated challenge within the workflow of ensemble
docking, but is directly related to the evaluation of docking results. Depending on
the strategy followed for structure selection docking results will either be a single
hit list or a collection of several hit lists. Subsequent pooling strategies for multiple
hit lists depend on the principal approach employed for selection of receptor representation.
15.2.2 Molecular Docking
15.2.2.1 Docking Algorithms and Scoring Functions
Molecular docking can be considered as one of the major pillars of structure-based
drug design. Docking flexible ligands, mostly small organic molecules, to a specific site of a receptor representing a rigid bio-molecular structure is an established
procedure [97, 98, 99, 100, 101]. Therefore, there is a critical awareness for the
potential as well as the limitations of this technique [92, 102]. Well accepted but still
open challenges are the inclusion of backbone flexibility (which is subject of this
overview), treatment of solvent and solvent-related effects [49], and quantitative