Recently an extensive benchmark study was carried out, evalu-
ating most of the currently available protein–protein docking pro-
tocols in respect to their ability to correctly predict transmembrane
dimer structures [16]. The authors studied eight protein–protein
docking tools, i.e., ZDOCK, ClusPro v.1.0, HEX, GRAMM-X
v.1.2.0, PatchDock (version beta 1.3), SymmDock (version beta
1.0), and HADDOCK. They used blind docking; only in case of
HADDOCK they indicated a pair of residues for interaction as only
such an option was available. They selected multimeric transmem-
brane proteins with known crystal structure deposited in PDB
database. In case of proteins which are for example pentamers
they divided a complex into two parts for docking or applied a
symmetric docking, i.e., a construction of a pentamer model based
on the structure of a monomer. In all the experiments they
obtained ten models that were characterized with B_RMSD (the
lowest RMSD in comparison to the crystal structure) and
A_RMSD (the average RMSD in comparison to the crystal struc-
ture). In addition, they determined CAPRI parameters and struc-
tural parameters, such as complex surface area, interface area, and
polar and hydrophobic contributions to complex surface area and
interface area. The analysis of results regarding B_RMSD indicated
that the best docking results were obtained with GRAMM-X
(median 0.27 A ̊) and ZDOCK (median 4.20 A ̊). The values of
median of A_RMSD were similar for all the studied tools
(11–13 A ̊). It can be concluded that best protein–protein docking
tools result in a few correct models which need to be separated
from a great number of incorrect models. Thus, it is not problem-
atic to obtain a correct model but it is a challenge to select scoring
functions for the obtained population of models which place the
correct models on the top of the ranking list (seeNote 3). The
encouraging results obtained using GRAMM-X may be illustrated
by the fact that it was possible to reconstruct 9 of 12 studied
complexes using this tool. As a comparison, the next successful
tools, i.e., ZDOCK, HADDOCK, ClusPro (symmetric docking),
and HEX were able to reconstruct four correct models only.
Importantly, only one correct model was obtained with ClusPro
using unsymmetrical docking. The main advantage of GRAMM-X
over other tools may be connected with the presence of evolution-
ary conservation score (for interface) in a scoring function in this
tool. The authors also analyzed structural features of transmem-
brane proteins which facilitate or hamper the application of pro-
tein–protein docking approach for generating reliable models. As it
can be expected, it is easier to model transmembrane protein
complexes with a large interface rich in structurally complementary
cavities. This explains the lack of success in application of protein–
protein docking technique to GPCR dimers. It is also well-known
that if the complex formation is accompanied by a significant
conformational change, protein–protein docking approach is not296 Agnieszka A. Kaczor et al.