Docking Predictions of Protein-Protein Interactions and Their Assessment: The CAPRI Experiment - Identification of Ligand Binding Site and Protein-Protein Interaction Area - page 86

5.1 Introduction

The specific recognition between two proteins is the physical process that governs
the construction of the macromolecular machines and assemblies which carry out
most biological functions in cells and living organisms. Ubiquitous and essential to
life, protein-protein recognition has in recent years become a major subject of study
in post-genomic molecular biology, biochemistry, structural biology, and biophysics.
When structural data are available, it can also be approached computationally, by
docking simulations in which a protein-protein complex is assembled from the
component structures. We relate here how protein-protein docking was attempted in
the early 1970s, preceding small molecule docking at a time where very few pro-
teins had a known three-dimensional structure, and how it developed into a family
of novel algorithms after 1990. Since then, docking algorithms have turned into
structural prediction procedures, and their reliability has been tested in the CAPRI
blind prediction experiment. An outcome of the test was that the initial model of
recognition in which the proteins bind as rigid bodies, progressively evolved into
one of flexible recognition. The new paradigm takes into account the structure
changes that may accompany the association reaction, and offers estimates of their
effect on the stability of the assembly that the reaction produces, and on the
specificity of the recognition process.

5.2 An Early History of Protein Docking

The first attempt to model the self-assembly of two proteins concerned trypsin
and the bovine trypsin pancreatic inhibitor (BPTI). David Blow of Cambridge,
UK, and Robert Huber of Martinsried, Germany, respectively authors of the
a-chymotrypsin and BPTI X-ray structures, teamed to build an atomic model of the
trypsin/BPTI complex. Their paper (Blow et al. 1972 ) does not say how they did it,
only that “when a model of the relevant part of the inhibitor was compared with the
active site of a-chymotrypsin, it was evident that only one mode of binding was pos-
sible”. At the time, “model” meant a physical wire model, not one a computer could
handle, and no atomic coordinates of the complex remain to assess its accuracy.
Beddell et al. ( 1976 ) still used a wire model to do molecular modeling at the Wellcome
Research Laboratories in Kent, UK. They engineered biphenyl compounds to bind at
the DPG (2,3-diphospho-glycerate) site of hemoglobin. Some of the compounds did,
and they had the predicted effects on oxygen binding, possibly the first success of
structure-based drug design. While the Wellcome scientists had access to hemoglobin
atomic coordinates from Pr. Max Perutz, their paper says that “a more accurate
representation was needed”, and they chose to build a wire model. Their designs
were based on interactions predicted from that model, not computation.
Nevertheless, Perutz' hemoglobin coordinates had already been used to do
molecular modeling in the computer, and more specifically, to dock proteins
together. Pr. Cyrus Levinthal of Columbia University, New York, had devised an