Chemistry Reference
In-Depth Information
results in a proportional increase in computational time. Also, the results may be
ambiguous, since there may be several equally valid ligand poses for each different
protein conformation. This is especially apparent in virtual screening approaches
where enrichment factors suffer when docking to multiple structures (please see
Sect. 3.4
). This is likely due to an increase in the number of false positives among
the top hits [
126
]. Ensemble docking is an alternative to docking multiple structures
that removes the ambiguity [
118
]. All the protein structures from the ensemble are
superimposed in order to generate an average structure or an average receptor grid.
The docking is then performed against the average structure or average receptor
grid (Fig.
5
). The ensemble docking approach allows for a single docking at a
significantly lower computational cost; however, it may suffer from accuracy
problems if the ensemble is biased towards the unbound form of the protein.
Effectively, a biased ensemble may negate the goal of incorporating protein
flexibility if it represents a single conformation.
3.4 Virtual Screening and Assessment
Using molecular docking to identify lead candidates is an attractive approach for
the pharmaceutical industry; it allows for the rapid evaluation of millions of
chemical compounds while using minimal resources compared to traditional
HTS. The process by which molecular docking is used to rank compounds within
a library based on a predicted binding affinity is known as virtual screening [
127
,
128
]. The potential benefit to drug discovery has inspired the development and
evaluation of numerous virtual screening methodologies. A virtual screen requires a
balance between optimizing speed and maximizing accuracy. Specifically, the goal
of a drug discovery virtual screen is the rapid and efficient separation of a small
subset of active compounds from a relatively large random library of inactive
compounds. Unfortunately, determining the effectiveness of a specific virtual
screening process is challenging, where independent evaluators routinely generate
inconsistent results [
87
,
129
-
131
].
The ambiguous nature of the results from a virtual screen requires additional
methods to evaluate its success. Typically, a virtual screening process is evaluated
against a protein target with a set of known binders. Assessing the performance of
a virtual screen is primarily based on the accuracy of the predicted ligand pose and
binding affinity. The correct binding pose is often evaluated by calculating the
RMSD between the docked and experimental ligand structures. The evaluation of
binding affinity is typically based on the accurate ranking of known binders instead
of the absolute scores because of the known limitations with calculating a binding
energy. Other modes of performance assessment involve evaluating enrichment and
generating diverse hit lists.
In a virtual screening protocol, every compound in a library (
N
tot
)isdockedto
the protein and a corresponding binding score is calculated. The binding score for
the ligand's best docked pose is used to rank the ligand relative to the entire library.
