504
S. von Grafenstein et al.
15.1.4 The Ensemble Docking Workflow
Docking to multiple receptor conformations (sometimes abbreviated MRC [8]) can
be applied on a set of conformations from any source. Experimentally different Xray conformations or NMR solution structures can represent the same biomolecular
system. NMR solution structures and conformations sampled during an MD simulation represent a valid thermodynamic ensemble, e.g., a canonical ensemble, for
simulations with constant number of particles, volume, and temperature. Although,
a collection of X-ray structures is not a valid thermodynamic ensemble, docking to
multiple X-ray structures is sometimes also termed ensemble docking.
The idea to combine molecular dynamics simulations and molecular docking is
an obvious way to include protein flexibility within docking if flexibility is not covered within experimental structures [22]. Here, we describe the concept of ensemble
docking to an MD-based ensemble by following the workflow and present recent
interesting examples applying this technique. The principle workflow is intuitive:
First, a structural ensemble of the receptor is generated via an MD simulation. Second, molecular docking into one or multiple representations of the ensemble is performed. Finally, the docking results for multiple representations are pooled into a
single score for each compound.
Several practical questions arise along this ensemble docking workflow where
different strategies can be followed. Moreover, we will discuss that they are not
only practical questions but are directly related to the underlying theory and should
be considered regarding the physical interpretation of final results.
• How to setup the MD simulation?
• How to select the right representation of the ensemble for docking?
• How to score given multiple hit lists?
At each step, we tried to systematically cover alternative approaches followed in
literature. Multiple branching points introduce a lot of variables in the workflow
making it hard to compare test cases from literature directly with each other. Therefore, we set up a test case where we can consistently address individual questions
and systematically show the influence of distinct strategies.
We will focus on influenza neuraminidase, an established drug target, where
computational techniques were already successfully applied in the development of
established inhibitors [23]. Influenza neuraminidase is a classical target to investigate and visualize the adaption of side chain orientation upon binding of alternative
ligands [24]. Additionally, backbone flexibility is an issue for ligand recognition
[25] and several studies applied molecular dynamic simulations in combination
with molecular docking for the investigation of novel lead structures [26, 27].
Data presented here are extracted from all-atom simulations of influenza neuraminidase with an explicit water model and represent a sampling time of 30 ns
(Fig. 15.1). The starting structures are three alternative X-ray structures of neuraminidase of recent forms of influenza A H1N1 without an inhibitor (Fig. 15.1a),
bound to the inhibitor zanamivir (compound 1; Fig. 15.1b) and an active site
S. von Grafenstein et al.
15.1.4 The Ensemble Docking Workflow
Docking to multiple receptor conformations (sometimes abbreviated MRC [8]) can
be applied on a set of conformations from any source. Experimentally different Xray conformations or NMR solution structures can represent the same biomolecular
system. NMR solution structures and conformations sampled during an MD simulation represent a valid thermodynamic ensemble, e.g., a canonical ensemble, for
simulations with constant number of particles, volume, and temperature. Although,
a collection of X-ray structures is not a valid thermodynamic ensemble, docking to
multiple X-ray structures is sometimes also termed ensemble docking.
The idea to combine molecular dynamics simulations and molecular docking is
an obvious way to include protein flexibility within docking if flexibility is not covered within experimental structures [22]. Here, we describe the concept of ensemble
docking to an MD-based ensemble by following the workflow and present recent
interesting examples applying this technique. The principle workflow is intuitive:
First, a structural ensemble of the receptor is generated via an MD simulation. Second, molecular docking into one or multiple representations of the ensemble is performed. Finally, the docking results for multiple representations are pooled into a
single score for each compound.
Several practical questions arise along this ensemble docking workflow where
different strategies can be followed. Moreover, we will discuss that they are not
only practical questions but are directly related to the underlying theory and should
be considered regarding the physical interpretation of final results.
• How to setup the MD simulation?
• How to select the right representation of the ensemble for docking?
• How to score given multiple hit lists?
At each step, we tried to systematically cover alternative approaches followed in
literature. Multiple branching points introduce a lot of variables in the workflow
making it hard to compare test cases from literature directly with each other. Therefore, we set up a test case where we can consistently address individual questions
and systematically show the influence of distinct strategies.
We will focus on influenza neuraminidase, an established drug target, where
computational techniques were already successfully applied in the development of
established inhibitors [23]. Influenza neuraminidase is a classical target to investigate and visualize the adaption of side chain orientation upon binding of alternative
ligands [24]. Additionally, backbone flexibility is an issue for ligand recognition
[25] and several studies applied molecular dynamic simulations in combination
with molecular docking for the investigation of novel lead structures [26, 27].
Data presented here are extracted from all-atom simulations of influenza neuraminidase with an explicit water model and represent a sampling time of 30 ns
(Fig. 15.1). The starting structures are three alternative X-ray structures of neuraminidase of recent forms of influenza A H1N1 without an inhibitor (Fig. 15.1a),
bound to the inhibitor zanamivir (compound 1; Fig. 15.1b) and an active site