Docking Predictions of Protein-Protein Interactions and Their Assessment: The CAPRI Experiment - Identification of Ligand Binding Site and Protein-Protein Interaction Area - page 94

It generated lively discussions on line through the CAPRI Website hosted by the
European Bioinformatics Institute (Hinxton, UK), and face-to-face during assess-
ment meetings that took place at regular intervals. Moreover, CAPRI is at the ori-
gin of most of the progress seen in the last 10 years. The new scoring functions, the
methods to model conformation changes and the flexible docking procedures
developed since 2001, owe much to the experiment. Docking searches based on
simulated annealing and molecular dynamics have been adapted to reproduce loop
movements or the rotation of a structural domain about a pre-defined hinge, with
good results on some CAPRI targets (Janin 2005 ; Lensink et al. 2007 ; Lensink and
Wodak 2010 ). However, a more general solution to the problem of flexible docking
is to generate conformers of the two components prior to the search, and assemble
them pairwise (Grünberg et al. 2004 ; Bonvin 2006 ; Lesk and Sternberg 2008 ;
Dobbins et al. 2008 ; Ritchie 2008 ; Zacharias 2010 ). A recent, and valuable, appli-
cation of the method is implemented in the MultiFit server (Table 5.1 ); it allows
multiple conformers generated from a X-ray structure to be fitted in electron
microscopy images that have a much lower resolution, but may display significant
conformation changes relative to the atomic model (Tjioe et al. 2011 ) .
The different approaches to the problem of conformation changes in docking
correspond to different possible recognition mechanisms. Rigid-body docking mimics
the specific recognition between two proteins that bear complementary surfaces,
ready to interact when they come into contact. Monte-Carlo searches with variable
dihedral angles simulate an induced fit mechanism in which the components first
make a low-stability, low-specificity contact, and then adjust their conformation to
optimize their interaction. Docking conformers pairwise closely reproduces con-
former selection, an alternative to induced fit. In this mechanism, a minority of the
molecules have the conformation that allows rigid-body recognition to start with,
and it is the formation of a complex that causes the equilibrium to shift (Grünberg
et al. 2004, 2006 ) .

5.5.2 Scoring in CAPRI

Docking conformers pairwise multiplies the number of searches, and this is practical
only with a fast docking algorithm (Schneidman-Duhovny et al. 2005 ; Mashiach
et al. 2010 ). In addition to being computationally demanding, the method generates
a great number of false positives, and puts a heavy load on the scoring functions.
In recent years, CAPRI has been adapted to assess scoring separately from docking.
The scoring part of the experiments operates in this way: after a prediction round is
completed, the predictor groups are asked to upload a hundred or so of their models,
which are merged into a file that may contain a thousand models, issued from ten or
more different procedures. The scorer groups download the file and rank the whole
set, to make their own ten-model submission. In several cases, target T37 of Fig. 5.1
for instance, the scorers' submissions contained more accurate models than the