512
S. von Grafenstein et al.
can be integrated in the docking algorithms as additional degree of freedom [63, 65,
66]. Zaccharias integrated the MD-derived collective variables [63] in a new docking algorithm and later NMA-derived collective variables and side chain flexibility
were combined [65].
Alternative sampling strategies use Monte Carlo (MC) algorithms to sample
alternative protein structures. Here, we will present two cases where active site
backbone flexibility is integrated in presence of a ligand. The IREDA algorithm
generates multiple pocket conformations by an algorithm reorienting the co-crystallized ligand in the pocket (seeding), minimizing the resulting complexes and finally optimizing the energy with a global MC protocol [70]. During the optimization sampling of protein flexibility is enhanced with random perturbation of torsion
angles around the binding site and small displacement and rotation of the ligand.
The procedure uses the internal coordinate system of the ICM docking protocol and
the generated structures are evaluated using a Poisson solvation energy term [70].
For a set of kinases the novel generated structures perform equally well as multiple
X-ray structures in cross-docking experiments and for the discrimination of active
from inactive ligands. Moreover, novel binding pockets were not overoptimized for
the ligand used for structure generation and allowed near-native poses for alternative ligands.
In contrast, ROSETTALIGAND docking uses MC sampling in presence of the
ligand to be docked [71]. The algorithm presented in this study is improved over the
first published ROSETTALIGAND algorithm [72] by integrating backbone receptor flexibility and additionally, a more exhaustive sampling of ligand flexibility.
The presented study used about 5000 trajectories for each ligand conformer limiting its large scale application. A potential reduction of this extensive sampling is
discussed by the authors. Still, the protocol is highly parallel and therefore scales
with the number of used CPUs. Each trajectory consists of three phases: first, a
coarse-grained placement of the ligand is performed and evaluated for shape complementarity focusing only on receptor backbone and C-β atoms; second, a MCbased minimization scheme is repeated for six cycles exploring side chain rotamer
libraries, ligand positions and torsions; third, a final minimization includes backbone flexibility. The energy is evaluated with the all-atom ROSETTA scoring function for the second and third step with soft van-der-Waals repulsion in the second
step. The bench–marking results show a clear improvement in comparison to the
previous ROSETTALIGAND implementation indicating the beneficial effect of
integrating backbone flexibility in terms of identification of the near-native ligand
poses for the ROSETTALIGAND test cases [72] and the Astex test set. However,
for a set of urokinase structures inclusion of backbone flexibility lead to generation
of highly ranked but non-native poses indicating an increased rate of false positive
hits by the additional degrees of freedom [71].
15.2.1.4 Representation of the Structural Ensemble
Choosing the optimal representation of a collection of conformers (ensemble) is a
crucial step independent of the origin of alternative conformations. Therefore, most
S. von Grafenstein et al.
can be integrated in the docking algorithms as additional degree of freedom [63, 65,
66]. Zaccharias integrated the MD-derived collective variables [63] in a new docking algorithm and later NMA-derived collective variables and side chain flexibility
were combined [65].
Alternative sampling strategies use Monte Carlo (MC) algorithms to sample
alternative protein structures. Here, we will present two cases where active site
backbone flexibility is integrated in presence of a ligand. The IREDA algorithm
generates multiple pocket conformations by an algorithm reorienting the co-crystallized ligand in the pocket (seeding), minimizing the resulting complexes and finally optimizing the energy with a global MC protocol [70]. During the optimization sampling of protein flexibility is enhanced with random perturbation of torsion
angles around the binding site and small displacement and rotation of the ligand.
The procedure uses the internal coordinate system of the ICM docking protocol and
the generated structures are evaluated using a Poisson solvation energy term [70].
For a set of kinases the novel generated structures perform equally well as multiple
X-ray structures in cross-docking experiments and for the discrimination of active
from inactive ligands. Moreover, novel binding pockets were not overoptimized for
the ligand used for structure generation and allowed near-native poses for alternative ligands.
In contrast, ROSETTALIGAND docking uses MC sampling in presence of the
ligand to be docked [71]. The algorithm presented in this study is improved over the
first published ROSETTALIGAND algorithm [72] by integrating backbone receptor flexibility and additionally, a more exhaustive sampling of ligand flexibility.
The presented study used about 5000 trajectories for each ligand conformer limiting its large scale application. A potential reduction of this extensive sampling is
discussed by the authors. Still, the protocol is highly parallel and therefore scales
with the number of used CPUs. Each trajectory consists of three phases: first, a
coarse-grained placement of the ligand is performed and evaluated for shape complementarity focusing only on receptor backbone and C-β atoms; second, a MCbased minimization scheme is repeated for six cycles exploring side chain rotamer
libraries, ligand positions and torsions; third, a final minimization includes backbone flexibility. The energy is evaluated with the all-atom ROSETTA scoring function for the second and third step with soft van-der-Waals repulsion in the second
step. The bench–marking results show a clear improvement in comparison to the
previous ROSETTALIGAND implementation indicating the beneficial effect of
integrating backbone flexibility in terms of identification of the near-native ligand
poses for the ROSETTALIGAND test cases [72] and the Astex test set. However,
for a set of urokinase structures inclusion of backbone flexibility lead to generation
of highly ranked but non-native poses indicating an increased rate of false positive
hits by the additional degrees of freedom [71].
15.2.1.4 Representation of the Structural Ensemble
Choosing the optimal representation of a collection of conformers (ensemble) is a
crucial step independent of the origin of alternative conformations. Therefore, most