Small Molecule Docking from Theoretical Structural Models - Computational Modeling of Biological Systems - page 79

cation- contacts, special metal-ligand interactions, presence of buried waters in
the binding cavity, and many others. All these different terms are weighted using
parameters that are fitted against empirical data. As discussed above, different
programs offer the user the possibility of using family specific scoring functions and
to incorporate his/her own scoring functions. However, the large number of available
scoring functions has generated an obvious confusion in the users community and
has driven to the popularization of strategies based on consensus or meta-scoring
functions. Future work needs to be done by the community to order this explosion
of different scoring strategies.
Flexibility is treated at different levels by various programs. Ligands with poten-
tial drug-like properties tend to be small and moderately flexible, which facilitates
the determination of the optimum docking conformation by different methods such
as energy minimization, Monte Carlo, genetic algorithms, molecular dynamics,
and many others. The complexity here arises from the need to determine which
is the optimum geometry in solution [ 6 ]. As noted above, the incorporation of the
protein flexibility is much more difficult due to the large number of protein degrees
of freedom, and none “final” algorithm has been yet developed. Many programs
allow the user to refine a reduced number of residues in the protein—generally
limited to side chains—by using rotamer libraries [ 29 ], Monte Carlo [ 30 ], or
restrained molecular dynamics [ 31 ]. Nevertheless, one of the most popular strategies
consists in the “ensemble” docking approach, which assumes that the effect of target
flexibility in docking can be represented by using a Boltzmann ensemble of confor-
mations for the protein instead of just a single rigid structure. Different methods
for generating ensembles have been proposed, including molecular dynamics from
a known experimental structure of the target [ 32 , 33 ], crystallographic (X-ray) [ 34 -
37 ], and spectroscopic (NMR) [ 38 , 39 ]-derived structures.
A common feature in most descriptions of new docking methods is the claim
that it is more accurate than the competitors. In our experience, the performance of
docking algorithms changes in each version and depends quite significantly on the
nature of the problem and the skills of the modeler running the project, factors that
hinder the validity of the conclusions derived from blind test experiments [ 40 ]. An
estimate of the market share taken by the different docking algorithms is also dif-
ficult to determine, particularly in a scenario of site-licenses, cost-related decisions
in the selection of docking engines and where publication is not often a priority.
However, a simple analysis of the literature (ISI CITATION MANAGER) in 2009
reveals that the market is quite equally divided among different codes (see Fig. 2 ).

3 Scenario for Docking Use

The literature is full of examples of use of docking algorithms in drug design
procedures, and the documentation accompanying the different computer programs
illustrates many examples where docking has been crucial to derive significant
results. Even though most docking studies are done inside pharmaceutical industries