Chemistry Reference
In-Depth Information
A road to improvements for many state-of-the-art docking programs is the
accuracy of the scoring functions (differentiation between active and nonactive
ligands). Many scoring functions can be influenced by physical properties of the
ligands. This bias involving physical properties can cause ligands with specific
physical properties to incorrectly score better than others. Normalizing molecular
docking rankings using virtually generated decoys were recently reported [818].
In 2012 a compendium of papers was published in volume 26, issue 6 of JCAMD.
This work focused on most widely used docking programs which were evaluated
for pose prediction/VS performance/strategies for improvement.
In 2013 Yuriev
et al.
[342] published their review on latest developments in
molecular docking. There are established programs as well as new docking
programs and their innovations (proposed to deliver improvements to address
shortcomings of existing methods). There is not however, an unambiguous
confirmation of claims better addressed by benchmarking studies.
Comparisons of different programs are generally performed using scoring and
docking to identify best poses and binding strengths of ligands in binding cavities.
No consistency was observed in terms for which scoring/programs/functions
performed better. The task is conflicting and not straightforward. This is due to
the types of ligands, targets, quality of target structures, practitioner's level of
experience, preparation of ligands, binding sites as well as other differences in
protocol.
For consistent objective comparisons development of appropriate test sets/decoy
sets are necessary. As expected, the main conclusions support the fact that pose
prediction in contrast to affinity estimation, is performed well, by existing
programs. For specific target types, each program performs better than the other.
This is probably due to poor scoring across the board. Good pose prediction does
not correlate necessarily well with good scoring.
Challenges remain regarding prediction of binding affinities, effective scoring and
proper sampling of ligands and receptors. When available, receptor flexibility is
still far from perfect. Side chains and backbones should also be considered. It is
