able to consider this change in conformation (seeNote 2). The
obtained results indicate that protein–protein docking should be
used with great care when applied to modeling transmembrane
protein complexes. Moreover, the main problem is to select a
correct model from a large population of obtained models. It can
be facilitated by taking into consideration the degree of evolution-
ary conservation of the interface and surface roughness.
One of the issues that prohibit protocols optimized for aque-
ousproteinstobeusedforGPCRsisscoring.Whencarryingout
protein–protein docking, desolvation energy is an important
scoring factor in many PPD methods. However, desolvation
energy of aqueous proteins is an estimate of the energetic effect
of residues which go out of the aqueous solution when engaging
in protein–protein contacts [77].Thisparameterisinaccurate
formembraneproteinsastheymainly desolvate from a hydro-
phobic membrane environment instead of an aqueous solution
(seeNote 3).
Currently there exists a protocol that aims specifically at gen-
erating accurate GPCR oligomer structures [79]. The approach
generates 144 input structures by rotating each monomer by an
increment of 30, and submits those complexes to docking by
Rosetta [80]. Afterward each complex is scored based on an exten-
sive consensus approach tailored toward GPCR proteins that incor-
porates 11 scoring factors. The protocol provided satisfactory
predictions for the structure of GPCR dimers available at the
time. This protocol was used to model the dopamine D 2 receptor
homodimer [81] resulting in asymmetric dimer model with
TM1-TM2-TM4-TM5 interface [82]. To our knowledge, cur-
rently there are no other protocols aimed specifically at GPCRs,
although in recent years a few other protocols aimed at membrane
proteins have been developed.
The membrane-protein version of DOCK/PIERR [83] scores
generated poses based on empirically derived residue contact
potentials and further rescores them using a membrane protein-
specific energy function. The rescoring function employs predicted
energy costs of residue transfers between the solute and membrane,
obtained from molecular dynamics experiments [84].
Transmembrane protein complexes are often modeled using
RosettaDock [5, 17, 80] which is a multiscale docking algorithm
based on the Monte Carlo (MC) method. This approach which
incorporates both a low resolution, centroid-mode, coarse-grain
stage and a high resolution, all-atom refinement stage that opti-
mizes both rigid-body orientation and side-chain
conformation [79].
A newly developed Rosetta framework for membrane proteins
(Rosetta MP) includes a protocol for protein docking [85]. It
consists of a prepacking step to generate a starting structure
(in which the proteins are separated by a distance, optimized by
Protein-Protein Docking 297