Chemistry Reference
In-Depth Information
clash interactions “softer.” This essentially helps minimize the effect of small
structural variations that may occur during ligand binding.
While the above scoring methods are generally useful in describing protein-
ligand interactions, the simplifications used in each approach limits the overall
accuracy in predicting the correct docked ligand pose [
113
,
114
]. The major
weakness of most docking programs has been shown to be the scoring function.
One approach to compensate for this deficiency is to use a consensus score from
a combination of scoring functions to rescore a docked pose. Consensus scoring
[
31
,
115
] has been shown in several examples to improve docking results compared
to a single scoring function. However, like individual scoring functions, the
improvement is not consistent and the proper choice of scoring functions to
calculate a consensus score is typically based on trial and error.
3.3 Protein Flexibility
Proteins are inherently flexible and undergo a range of motions over different time
scales, and thus the use of rigid protein structures by molecular docking is prob-
lematic [
116
,
117
]. This is especially troublesome for therapeutic targets where
only an apo-structure is available. Conformational changes upon ligand-binding
may range from small perturbations in side chain conformation at the site of ligand
binding to large rearrangements of the entire protein structure. Not accounting for
such structural changes during ligand docking can drastically alter the ability to
identify reliable protein-ligand models correctly [
118
-
122
]. Conversely, attempting
to dock a large library of flexible ligands to a completely flexible protein structure
using molecular dynamics is too computationally expensive to be practical.
Several approaches to “solve” the protein flexibility problem have been
explored. The first generally applicable approach utilized soft docking in the
scoring function, which reduces the van der Waals repulsion terms in the empirical
scoring function [
123
,
124
]. This allows for some overlap between ligand and
protein atoms. While this approach is simple and fast, it can only accommodate
very small changes in side chain conformations. Other approaches attempt to
implement protein structural changes into the docking process. For example, a
library of side chain rotamers for residues only in the ligand binding site is routinely
used [
40
,
125
]. This dramatically reduces the number of active rotatable bonds
during the docking process and has a lower computational cost compared to
molecular dynamics. However, the inclusion of a library of rotamers in the docking
protocol
is significantly slower than rigid protein docking. Furthermore,
the
approach is limited to local side chain conformational changes.
The most common docking technique that attempts to account for protein
flexibility uses multiple protein structures. The ensemble of structures is expected
to represent the range of conformations sampled by the protein and has the benefit
of being able to evaluate both small and large conformational changes. The
molecular docking is repeated for each individual protein conformation, which
