design of vaccines or anti-infective drugs. Protein–protein interac-
tions are thus responsible for various functions of the cell. The
nature of those interactions can be better explained by applying
three-dimensional structures of protein–protein complexes and
binding affinity data [3]. Understanding the physical and structural
principles which govern proteins binding is a challenging and only
partially resolved problem in structural bioinformatics [4].
Targeting protein–protein interactions by small-molecule modu-
lators has received a considerable attention as a drug design approach,
conditioned, however, by the knowledge of 3D structure of a protein
complex. [5]. Unfortunately, design of compounds targeting protein
complexes has been hampered for a long time by the typical large and
flat nature of protein interaction surfaces, often missing clear pockets
for potential modulators [2]. The discovery of so-called “hot spots” in
PPI interfaces which are well-defined regions responsible for most of
the protein–protein binding energy has enabled the design and dis-
covery of small molecule modulators, including orthosteric PPI inhi-
bitors, stabilizers, and allosteric modulators [2].
In order to design PPI modulators, the 3D structure of the
protein–protein complex has to be known from experimental or
computational studies. As the number of X-ray-resolved protein–-
protein complexes is limited in available databases, computational
approaches are a method of choice [6–8] when obtaining a reliable
structure of the complex. In particular, protein–protein docking
(PPD), which is not as resource-demanding as molecular dynamics,
is commonly used for predicting the structures of protein com-
plexes. While small-molecular docking is one of the most com-
monly applied molecular modeling methods (starting from the
DOCK method published in 1982 by Kuntz et al. [9]), it should
be stressed, however, that protein–protein docking is as old method
(or even older) as traditional small-molecular docking [8], as the
first true example of protein–protein docking originates to as early
as 1970s [10]. Unfortunately, PPD is often much more problem-
atic than docking of small molecules. One of the main reasons is the
dynamic nature of proteins. While small molecules have a limited
number of degrees of freedom, and some of them are rigid and
well-defined, a typical protein structure undergoes constant fluc-
tuations to a remarkable degree, and therefore should rather be
considered as an ensemble of structures. This raises the problem
with generating sufficient ensemble of conformations of docked
proteins. Moreover, while a small-molecule orthosteric binding site
can be reliably identified with various methods, identification of
protein interaction interfaces is a challenging task. In terms of pose
sampling, it is a bit easier when at least one of the partner is a
membrane-spanning protein, because then it is possible to restrict
the search for interaction interfaces to some protein surfaces (trans-
membrane regions when both partners are membrane-spanning
proteins, or extracellular/intracellular when the other protein is286 Agnieszka A. Kaczor et al.