528
S. von Grafenstein et al.
structures. Comparison to the experimental structure in the validation is a necessary
step in order to judge the value of the MD-derived ensemble. Moreover, different
performance of two methods should be tested for significance [122]. Okimoto et al.
evaluated their screening method using a combination of docking and subsequent
MM/PBSA calculation against classical docking and showed that the new method
performed better for 3 out of 4 test scenarios with a confidence of 95 % [124].
One major motivation for ensemble docking is the assumption that ligands with
an alternative scaffold might bind to another protein conformation. To include this
aspect the ligands for evaluation should ideally represent a collection of chemically
diverse ligands. A metric covering this aspect would be the chemotype enrichment
[45, 114, 121]. The definition of a chemotype is not trivial in this context. Considering for example the two molecules used in this test case, they might fall into the
same chemotype based on the sialic acid scaffold. However, they bind to different
conformations of neuraminidase. Ensembles of X-ray structures were able to score
more diverse ligands compared to single structures [114, 121].
The plethora of measures for the evaluation makes an objective comparison between different studies nearly unfeasible. The evaluation in respect to experimental
data is always limited by existing knowledge. For some targets the chemical space
of known inhibitors does not allow to include a measure of diversity enrichment.
Additionally, the data we refer to during the validation is subject of change with
increasing number of available X-ray structures and activity data.
Ultimately, to prove that a method is able to identify new binders the experimental verification upon application to virtual screening is the ideal proof of concept. Successful identification of novel inhibitors for the potential Trypanosoma
brucei drug target RNA editing ligase 1 by the relaxed-complex-scheme represents
such a verification [84, 91]. Additionally, for the alternative target UDB-galactose
4′- epimerase of this pathogen a hit compound was experimentally confirmed and
led to a series of inhibitors with low µM inhibition. Similar promising results were
obtained for neuraminidase and were patented based on the µM activity in a flourescence-based enzyme inhibition assay [91, 125].
15.2.2.5 Rescoring of Highly Ranked Docking Poses
Apart from rescoring with alternative scoring functions post-processing of docking
results can include subsequent MD simulations based on the docking pose(s). As
post-processing requires additional calculation time these calculations are normally
restricted to highly ranked poses. Besides of the classical techniques for free energy
estimation with the help of MD simulations, we will first discuss an alternative MDbased metric to evaluate the poses. Classically, the starting point of the protocols
is an X-ray structure of a ligand-protein complex or a docking pose from a rigid
docking protocol.
Proctor et al. evaluated multiple docking poses by the stability of the pose during a subsequent simulation [126]. They used non-classical discrete molecular dynamics simulations which provide a fast sampling based on multiple independent
S. von Grafenstein et al.
structures. Comparison to the experimental structure in the validation is a necessary
step in order to judge the value of the MD-derived ensemble. Moreover, different
performance of two methods should be tested for significance [122]. Okimoto et al.
evaluated their screening method using a combination of docking and subsequent
MM/PBSA calculation against classical docking and showed that the new method
performed better for 3 out of 4 test scenarios with a confidence of 95 % [124].
One major motivation for ensemble docking is the assumption that ligands with
an alternative scaffold might bind to another protein conformation. To include this
aspect the ligands for evaluation should ideally represent a collection of chemically
diverse ligands. A metric covering this aspect would be the chemotype enrichment
[45, 114, 121]. The definition of a chemotype is not trivial in this context. Considering for example the two molecules used in this test case, they might fall into the
same chemotype based on the sialic acid scaffold. However, they bind to different
conformations of neuraminidase. Ensembles of X-ray structures were able to score
more diverse ligands compared to single structures [114, 121].
The plethora of measures for the evaluation makes an objective comparison between different studies nearly unfeasible. The evaluation in respect to experimental
data is always limited by existing knowledge. For some targets the chemical space
of known inhibitors does not allow to include a measure of diversity enrichment.
Additionally, the data we refer to during the validation is subject of change with
increasing number of available X-ray structures and activity data.
Ultimately, to prove that a method is able to identify new binders the experimental verification upon application to virtual screening is the ideal proof of concept. Successful identification of novel inhibitors for the potential Trypanosoma
brucei drug target RNA editing ligase 1 by the relaxed-complex-scheme represents
such a verification [84, 91]. Additionally, for the alternative target UDB-galactose
4′- epimerase of this pathogen a hit compound was experimentally confirmed and
led to a series of inhibitors with low µM inhibition. Similar promising results were
obtained for neuraminidase and were patented based on the µM activity in a flourescence-based enzyme inhibition assay [91, 125].
15.2.2.5 Rescoring of Highly Ranked Docking Poses
Apart from rescoring with alternative scoring functions post-processing of docking
results can include subsequent MD simulations based on the docking pose(s). As
post-processing requires additional calculation time these calculations are normally
restricted to highly ranked poses. Besides of the classical techniques for free energy
estimation with the help of MD simulations, we will first discuss an alternative MDbased metric to evaluate the poses. Classically, the starting point of the protocols
is an X-ray structure of a ligand-protein complex or a docking pose from a rigid
docking protocol.
Proctor et al. evaluated multiple docking poses by the stability of the pose during a subsequent simulation [126]. They used non-classical discrete molecular dynamics simulations which provide a fast sampling based on multiple independent