530
S. von Grafenstein et al.
bound and free state gives an estimate of protein-ligand interactions and hence free
energy of binding. Targets with a wide and flexible binding pocket, like the promiscuous monooxygenase cytochrome P450 isoform 2D6, potentially allow alternative binding modes. To include this flexibility Stjernschantz et al. developed a
LIE model on different initial poses generated by docking for a series of thiourea
compounds [132]. Marelius et al. included the free energy of the LIE in comparison
to the results by scoring to illustrate the fluctuation over the simulation time for the
arabinose binding protein as for this target a LIE model had been already adapted
[48]. A problem of LIE is the choice of weighting parameters for the energy contributions. Depending on the system these coefficients need readjustments, hence
hindering the establishment of global models.
15.3 Potential of Ensemble Docking
We gave an overview on recent applications of ensemble docking with the focus on
MD-derived ensembles. Following the practical workflow it becomes clear that no
common recipe for a best practice has been established to date. At multiple stages
variability occurs and at each branching point the possible number of workflows
increases. This heterogeneity of the different applications limits the comparability
between them. Moreover, lack of consistent performance metrics and performance
comparison in respect to docking to single or multiple X-ray structures also limits
to quantify the benefit from ensemble docking to MD-derived ensembles. A systematic evaluation on a tailored benchmark set could allow to characterize the impact
of variability within the field and also to identify an optimal workflow. However,
considering the complexity of the ligand recognition process individual contributions depend on the system investigated. Therefore, we also suspect the “optimal”
workflow to be dependent on the system of interest. One of the most relevant points
of opposing concepts is the scoring of multiple poses. We expect that consideration
of thermodynamic concepts will aid to adapt molecular docking algorithms and
scoring functions on ensembles. The application of scoring functions and especially
the averaging of scoring values need a conscious handling and interpretation of
the resulting scores. These values might be valid to rank hit lists, which is the primary aim of scoring. However, interpretation in terms of thermodynamic properties
should be reflected carefully.
Despite of current absence of a clear answer to the question “How to?”, single
success stories and also attempts to benchmark the method provide a growing number of pieces of evidence to establish this method within virtual screening. MD
simulations or alternative computational sampling methods are an essential way
to investigate the target’s flexibility, especially if the flexibility is not covered by
experimental ensembles. The gained knowledge on the target gathered while considering a proper simulation setup or gained directly from the simulation data will
be of benefit for a structure-based drug discovery project anyway.
S. von Grafenstein et al.
bound and free state gives an estimate of protein-ligand interactions and hence free
energy of binding. Targets with a wide and flexible binding pocket, like the promiscuous monooxygenase cytochrome P450 isoform 2D6, potentially allow alternative binding modes. To include this flexibility Stjernschantz et al. developed a
LIE model on different initial poses generated by docking for a series of thiourea
compounds [132]. Marelius et al. included the free energy of the LIE in comparison
to the results by scoring to illustrate the fluctuation over the simulation time for the
arabinose binding protein as for this target a LIE model had been already adapted
[48]. A problem of LIE is the choice of weighting parameters for the energy contributions. Depending on the system these coefficients need readjustments, hence
hindering the establishment of global models.
15.3 Potential of Ensemble Docking
We gave an overview on recent applications of ensemble docking with the focus on
MD-derived ensembles. Following the practical workflow it becomes clear that no
common recipe for a best practice has been established to date. At multiple stages
variability occurs and at each branching point the possible number of workflows
increases. This heterogeneity of the different applications limits the comparability
between them. Moreover, lack of consistent performance metrics and performance
comparison in respect to docking to single or multiple X-ray structures also limits
to quantify the benefit from ensemble docking to MD-derived ensembles. A systematic evaluation on a tailored benchmark set could allow to characterize the impact
of variability within the field and also to identify an optimal workflow. However,
considering the complexity of the ligand recognition process individual contributions depend on the system investigated. Therefore, we also suspect the “optimal”
workflow to be dependent on the system of interest. One of the most relevant points
of opposing concepts is the scoring of multiple poses. We expect that consideration
of thermodynamic concepts will aid to adapt molecular docking algorithms and
scoring functions on ensembles. The application of scoring functions and especially
the averaging of scoring values need a conscious handling and interpretation of
the resulting scores. These values might be valid to rank hit lists, which is the primary aim of scoring. However, interpretation in terms of thermodynamic properties
should be reflected carefully.
Despite of current absence of a clear answer to the question “How to?”, single
success stories and also attempts to benchmark the method provide a growing number of pieces of evidence to establish this method within virtual screening. MD
simulations or alternative computational sampling methods are an essential way
to investigate the target’s flexibility, especially if the flexibility is not covered by
experimental ensembles. The gained knowledge on the target gathered while considering a proper simulation setup or gained directly from the simulation data will
be of benefit for a structure-based drug discovery project anyway.