Small Molecule Docking from Theoretical Structural Models - Computational Modeling of Biological Systems - page 76

and in fact some dirty drugs, 1 such as chlorpromazine, dextromethorphan, and ibo-
gaine exhibit desired pharmacological properties [ 2 ]. These considerations highlight
the tremendous difficulty of designing small molecules that both have satisfactory
ADME properties and the ability of interacting with a limited set of target proteins
with a high affinity, avoiding at the same time undesirable interactions with other
proteins. In this complex and challenging scenario, computer simulations emerge as
the basic tool to guide medicinal chemists during the drug discovery process.
Since early works in the 1980s, molecular docking has arised as a leading
simulation technique to facilitate the drug design. The traditional paradigm of
docking, known as rigid-body docking approach, assumes implicitly the Fisher's
lock-and-key model [ 3 ], and considers that the ligand-induced structural changes of
the protein are negligible [ 4 ]. However, drugs generally exhibit a certain degree of
flexibility, and the bioactive conformation might not be the most stable conformation
in solution [ 5 , 6 ]. This fact leads to the need of considering drug flexibility for a
successful docking simulation. Furthermore, analysis of the Protein Data Bank [ 7 ]
reveals that ligand binding can introduce non-negligible changes in protein structure
which often affect the binding site, raising tremendous difficulties for docking
techniques, especially in cases where structural changes are not only binding-
specific, but also drug-specific [ 8 ]. A second limitation in docking experiments
arises from the evaluation of the ligand-binding free energy. Free-energy simulation
techniques are expensive calculations that remain impractical for the evaluation
of large numbers of ligands [ 9 ]. Current docking strategies are based on the
combination of very fast functions, which intend to predict binding poses and rank
them by means of a more complex equation (the “scoring function”), which has been
parameterized to reproduce experimental binding data of protein-drug complexes
[ 10 ]. However, scoring functions implemented in docking programs make various
assumptions and simplifications, and do not fully account for all phenomena that
determine molecular recognition.
Despite all the challenges, the major practical limitation for docking procedures
does not emerge from technical uncertainties in the evaluation or scoring of
docking poses, but comes from the lack of experimentally solved protein structures.
Indeed, despite the massive effort focused in the experimental resolution of protein
structures, 2010 version of the PDB contains less than 4,000 unique human proteins,
while RefSeq [ 11 ] suggests the existence of nearly 100,000 human proteins, twice
or more if splicing variants are considered. Therefore, the current version of PDB
is covering only around 4% of the known human proteome [ 12 ]. This sequence-
structure gap becomes even larger if we consider proteins from virus, bacteria, or
other pathogens for which less amount of structural information exists.
The evaluation of the potential interactions of drugs with multiple targets is
severely limited if the analysis relies exclusively on experimentally solved struc-
tures. Fortunately, this limitation can be partially solved with the use of predicted
models of proteins as templates for docking (Fig. 1 ). In this chapter, we very briefly
1 Drugs that bind to several molecular targets or receptors, and therefore tend to have a wide range
of effects and possibly negative side effects.