variation that occurs during the formation of the ligand–receptor
complex is measured by the binding constant (Kd) and Gibbs free
energy (ΔGL)[18]. Scoring functions estimate these parameters
using three types of algorithms: (1) force-field-based, (2) empirical,
and (3) knowledge-based functions [19]. Those methods based on
force fields calculate the binding energy by summing bonded (bond
stretching, angle bending, and dihedral variation) and nonbonded
contributions (electrostatic and van der Waals interactions). Force-
field-based methods use ab initio calculations and the equations of
classical mechanics to estimate the contribution of each of these
terms to the total binding energy. Empirical scoring functions use
training sets consisting of protein–ligand complexes with known
binding affinities to generate statistical models. The terms used in
the derivation of the models consist of hydrogen-bonding, ionic
and non-polar interactions, as well as desolvation effects and entro-
pic contributions [20]. The accuracy of empirical scoring functions
depends directly on the quality of the data used to generate the
model. The other approach, knowledge-based scoring functions,
uses data from known ligand–receptor complexes to calculate pair-
wise energy potentials and generate a general equation [21]. These
potentials are constructed by taking into account the frequency
with which two different atoms are found to interact in a series of
ligand–receptor complexes. The different types of interactions are
classified and weighted according to their occurrence to generate
the final scoring function, which is a sum of these individual inter-
actions. In short, current molecular docking programs predict the
conformation of a small molecule within the target binding site
with reasonable accuracy, as confirmed by comparing complexes
predicted by different algorithms with their respective X-ray struc-
tures. The major limitation is the lack of a suitable function to
estimate desolvation contributions, entropic effects, and protein
flexibility with reasonable accuracy [22, 23].1.2 Virtual Screening Virtual screening is the use of fast and cost-effective computational
methods to identify potentially active molecules from virtual data-
bases [24]. Exploring virtual compound collections is one of the
most widespread strategies in drug discovery, and several pharma-
cological agents can trace their origins to virtual screening efforts
[2, 6]. Strategies that rely on docking compounds from libraries
against a given molecular target are termed structure-based virtual
screening (SBVS) and are the focus of this chapter [24]. In addition
to the prediction of likely binding conformations, SBVS offers a
convenient method of ranking the docked molecules using scoring
functions. This classification criterion can be used solely or in
combination with other procedures for selecting promising mole-
cules for experimental profiling.
SBVS strategies usually rely on the following procedures:
(1) molecular target selection and preparation, (2) compound
34 Ricardo N. dos Santos et al.