516
S. von Grafenstein et al.
inhibitor, compound 2, known to bind to the open 150-loop conformation of neuraminidase (see Fig. 15.1c). Each of the three different simulations were represented
by 300 periodically extracted snapshots and docking to in total 900 structures was
performed with the program GOLD using the knowledge-based fitness function
GOLDSCORE. As discussed before (Sect. 15.2.1.2) conformations from the ligandbound simulations are optimized for zanamivir and result in the most favorable
poses, reflected by high docking scores (Fig. 15.4a, c). The pose with the best score
represents a near-native structure (Fig. 15.4e). As expected a near-native pose of
compound 2 is generated with a conformation from the open simulation. The low
scoring value for this pose might be related to the conformation of the Arg371 side
chain in the receptor conformation which is unfavorable to establish the close contact to the ligand’s carboxylate group (Fig. 15.4g). Consequently, a higher score is
generated with a snapshot of the zanamivir-bound simulation showing a flipped
ligand orientation in comparison to the native-pose from the X-ray structure but
an optimal contact between Arg371 and the ligand (PDB code 3O9K) (Fig. 15.4g).
It was shown that the best scoring value does not necessarily represent a binding
pose close to the crystal structure [80]. Single values showed artificially high nonphysical binding energies generated for docking poses with clashes between ligand
and protein. Subsequent minimization or post-processing using MM/PBSA as alternative free energy estimation techniques yielded top ranking for the pose near to
the native X-ray structure [82]. Without further post-processing ranking of different
compounds by the arithmetic mean over the scoring profile is suggested and the
mean of scoring values was successfully applied in selecting a lead modification for
HIV protease inhibitors [82].
Concepts similar to the RCS are applied by other groups using alternative simulation protocols or docking algorithms [24, 86]. Most of them do not present applications but rather evaluate the technique on multiple test cases in comparison to
other virtual screening techniques or aim at the identification of useful parameters
for the workflow. For example, Paulsen et al. investigate how the pooling of docking scores impacted the ranking accuracy for two test cases: Candida albicans dihydrofolate reductase and influenza neuraminidase [86]. Handling of multiple scoring
values is a major question in ensemble docking, which we will address later (see
Sect. 15.2.2.3). An exemplary application of a similar method is the virtual screening study on the estrogen receptor predicting novel leads in silico, unfortunately
lacking experimental verification [85].
Successful virtual screening projects using the RCS are summarized in a paper
by Amaro et al [84]. However, for these studies an alternative method to select the
conformations from the ensemble was chosen as multiple receptor conformations
increase the docking time per compound linearly with the number of conformations
considered. Therefore, especially focusing on virtual screening application, where
the compounds to be docked can easily exceed 105, the number of considered snapshots has to be limited. In addition to the increase in docking time per compound,
more conformations demand a higher overhead time as each protein conformation
must be prepared separately for the docking run depending on the docking algorithm (e.g., grid generation in GLIDE).
S. von Grafenstein et al.
inhibitor, compound 2, known to bind to the open 150-loop conformation of neuraminidase (see Fig. 15.1c). Each of the three different simulations were represented
by 300 periodically extracted snapshots and docking to in total 900 structures was
performed with the program GOLD using the knowledge-based fitness function
GOLDSCORE. As discussed before (Sect. 15.2.1.2) conformations from the ligandbound simulations are optimized for zanamivir and result in the most favorable
poses, reflected by high docking scores (Fig. 15.4a, c). The pose with the best score
represents a near-native structure (Fig. 15.4e). As expected a near-native pose of
compound 2 is generated with a conformation from the open simulation. The low
scoring value for this pose might be related to the conformation of the Arg371 side
chain in the receptor conformation which is unfavorable to establish the close contact to the ligand’s carboxylate group (Fig. 15.4g). Consequently, a higher score is
generated with a snapshot of the zanamivir-bound simulation showing a flipped
ligand orientation in comparison to the native-pose from the X-ray structure but
an optimal contact between Arg371 and the ligand (PDB code 3O9K) (Fig. 15.4g).
It was shown that the best scoring value does not necessarily represent a binding
pose close to the crystal structure [80]. Single values showed artificially high nonphysical binding energies generated for docking poses with clashes between ligand
and protein. Subsequent minimization or post-processing using MM/PBSA as alternative free energy estimation techniques yielded top ranking for the pose near to
the native X-ray structure [82]. Without further post-processing ranking of different
compounds by the arithmetic mean over the scoring profile is suggested and the
mean of scoring values was successfully applied in selecting a lead modification for
HIV protease inhibitors [82].
Concepts similar to the RCS are applied by other groups using alternative simulation protocols or docking algorithms [24, 86]. Most of them do not present applications but rather evaluate the technique on multiple test cases in comparison to
other virtual screening techniques or aim at the identification of useful parameters
for the workflow. For example, Paulsen et al. investigate how the pooling of docking scores impacted the ranking accuracy for two test cases: Candida albicans dihydrofolate reductase and influenza neuraminidase [86]. Handling of multiple scoring
values is a major question in ensemble docking, which we will address later (see
Sect. 15.2.2.3). An exemplary application of a similar method is the virtual screening study on the estrogen receptor predicting novel leads in silico, unfortunately
lacking experimental verification [85].
Successful virtual screening projects using the RCS are summarized in a paper
by Amaro et al [84]. However, for these studies an alternative method to select the
conformations from the ensemble was chosen as multiple receptor conformations
increase the docking time per compound linearly with the number of conformations
considered. Therefore, especially focusing on virtual screening application, where
the compounds to be docked can easily exceed 105, the number of considered snapshots has to be limited. In addition to the increase in docking time per compound,
more conformations demand a higher overhead time as each protein conformation
must be prepared separately for the docking run depending on the docking algorithm (e.g., grid generation in GLIDE).