Biology Reference
In-Depth Information
libraries of model assemblies have been built in this way (Lu et al.
2003
; Stein et al.
2011
). Templates may also be selected on the basis of the local similarity of the
protein surfaces: two surfaces that have a similar geometry and similar physical-
chemical features may be expected to make similar interactions (Günther et al.
2007
; Keskin et al.
2008
), in which case the PDB may already be adequate to repre-
sent the diverse architectures observed in nature (Tuncbag et al.
2008
; Kundrotas
et al.
2012
). Here again, the quality of the models remains to be assessed.
5.4
Assessing Docking Predictions: The CAPRI Experiment
5.4.1
CAPRI
By the turn of the century, several docking algorithms had developed into full-
fledged prediction procedures (see reviews by Smith and Sternberg
2002
; Camacho
and Vajda
2002
; Halperin et al.
2002
). At that time, an entirely new field of applica-
tion opened, due to the structural genomics (or proteomics) initiatives that accom-
panied the completion of the human genome sequence. High-throughput X-ray and
NMR studies were going to determine the structure of thousands of new proteins
that would include the components of many binary or larger assemblies. Docking
procedures could in principle build models of these assemblies from the component
structures, but should we trust the results at all? The procedures had been thor-
oughly tested, but most of the unbound docking tests had been done on protease/
inhibitor or antigen/antibody complexes, the only ones for which the component
structures were available. How would docking perform on new, possibly very different,
systems, and how accurate would the models be?
These questions were discussed in Charleston, South Carolina, in June 2001, at
a meeting on Modeling Protein Interactions in Genomes organized by Pr. Sandor
Vajda and Ilya Vakser, and the conclusion was that a blind prediction experiment
should be organized (Vajda et al.
2002
). Named CAPRI (Critical Assessment of
PRedicted Interactions), the experiment was modeled after CASP (Critical
Assessment of Structural Predictions), an older experiment that tests methods to
predict a protein fold based on its amino acid sequence (Moult et al.
1995
) . The
targets of CAPRI would be protein-protein complexes, and the prediction start from
component structures taken from the Protein Data Bank. The predictors would dock
the components, and submit models to the CAPRI Website, to be assessed by com-
parison with a newly determined, but unpublished, experimental structure of the
complex (Janin et al.
2003
). A blind prediction of that sort had been done once
before, on a b-lactamase in complex with a protein inhibitor. Six participant groups
had submitted models of the complex that were close to the X-ray structure
(Strynadka et al.
1996
). Could that performance be reedited?
An answer came soon after the Charleston meeting. The first round of CAPRI,
held in the summer of 2001, had three targets, three complexes whose X-ray struc-
tures had just been determined by collaborators of mine, willing to help starting the
experiment. Two were viral antigen proteins in complex with monoclonal antibodies,
Search WWH ::
Custom Search