present approaches are able to consider at least protein side-chain
flexibility [8]. Flexibility is a critical condition of the quality of
protein–protein docking, as in the CAPRI experiment most of the
failures are connected with inaccurately predicting protein confor-
mational changes upon protein–protein interaction [8]. One possi-
bility to directly apply computationally efficient rigid docking
algorithms is to indirectly consider receptor flexibility by represent-
ing the receptor target as an ensemble of structures (e.g., from NMR
studies or molecular dynamics simulations) [17]. A limited number
of starting configurations (in case of knowledge of the binding sites)
can bemodeled by combination ofdockingwith molecular dynamics
or Monte Carlo simulations which enables full atomic flexibility or
flexibility restricted to relevant parts of the proteins during docking
[17]. In case of some protein complexes, not only partner adjust-
ment occurs during the process of binding, but also a global confor-
mational change takes place. In such a situation Elastic Network
Model (ENM) calculations (based on simple distance-dependent
springs between protein atoms) may be successfully used to address
the mobility of proteins around a stable state [17]. Another possibil-
ity is to apply soft collective normal mode directions as additional
variables during docking by energy minimization [17].3 Algorithms and Software for Protein–Protein Docking
Global docking programs that are used for ab initio docking,
sample the relative binding orientations or poses between interact-
ing proteins over all six translational and rotational degrees of
freedom without the necessity of knowledge about the binding
site [5]. Due to the high cost of the searching stage of docking,
an efficient search strategy is critical for good performance of global
docking software. In this context, the majority of available global
docking programs make use of FFT (Fast Fourier Transform) cor-
relation search algorithms (FTDock, GRAMM, MolFit, DOT,
ZDOCK, PIPER), while few global docking programs such as
ATTRACT, PatchDock, and SwarmDock are based on other
types of search strategies like randomized search or local shape
matching over the whole protein [5]. Because of the required
computational efficacy, some early docking programs used only
simple scoring functions including shape complementarity and/or
hydrophobic interactions (e.g., GRAMM or MolFit). More
advanced programs, such as PatchDock uses geometric hashing
for fast surface patch matching [5]. ZDOCK also employs an
advanced pairwise method to use efficiently shape complementarity.
Electrostatic interactions contribution was first introduced in Fit-
Dock and later became routine in such programs like MolFit, DOT,
HEX, and ATTRACT. The desolvation effects are also considered
in ZDOCK, FRODOCK, and SDOCK. A summary of main global
docking programs is given in Table1.288 Agnieszka A. Kaczor et al.