Docking Predictions of Protein-Protein Interactions and Their Assessment: The CAPRI Experiment - Identification of Ligand Binding Site and Protein-Protein Interaction Area - page 88

5.3 Protein-Protein Docking Algorithms

5.3.1 Bound vs. Unbound Docking

The hemoglobin simulation faced even more editorial skepticism than the trypsin/
BPTI study. By the time it got published (Janin and Wodak 1985 ) , small molecule
docking had come of age in the hands of Kuntz et al. ( 1982 ) , Goodford ( 1985 ) , and a
few others. Soon, it became an established procedure in drug design, while protein-
protein docking remained confidential for over a decade. Meanwhile, computers
became orders of magnitude faster, and crystallographers determined many new struc-
tures. The latter included a score of protease/inhibitor complexes, and the first antigen/
antibody complexes (Janin and Chothia 1990 ) . Cher fi ls et al. ( 1991 ) tested on those
complexes the Wodak-Janin algorithm, implemented as a simulating annealing proce-
dure to make the search more efficient. This allowed all six degrees of freedom to be
explored, and most importantly, “unbound” docking to be tested for the first time.
Unbound docking uses the atomic coordinates of the free proteins, bound docking,
coordinates taken from the complex. Bound docking ignores the conformation changes
that may accompany association, and it has no predictive value, since the solution
must be known in advance. The new study yielded native-like models of all the target
complexes, and a majority of those models scored near the top. However, there were
many false positives, especially with the unbound proteins, and it was evident that
other features than shape complementarity had to be taken into account to identify the
correct docking models among all the false positives.

5.3.2 Rigid-Body Docking

The early 1990s were a period of renewed interest in protein-protein docking.
Several new algorithms, all based on geometry and shape complementarity, were
published almost simultaneously. Connolly ( 1986 ) had devised a procedure in which
molecular surfaces were described by sets of discrete points; matching critical
points (holes and pits) of two surfaces assessed their complementarity, and this
could be used for docking. A related method of surface triangulation, independently
developed for “computer-vision” by Pr. Haim Wolfson of Tel Aviv University in
Israel, was implemented into a docking procedure through a very efficient geometric
hashing algorithm (Nussinov and Wolfson 1991 ; Norel et al. 1994 ) . In Berkeley,
California, Jiang and Kim ( 1991 ) designed a “cube representation” of proteins
specifically for docking. In that model, the surface of the proteins and their interior
volume are sampled on a cubic grid, and a docking pose is generated by matching
surface cubes while rejecting overlaps between volume cubes. Jiang and Kim made
a very important point: docking must be “soft” to allow for minor conformation
changes. The cube model, like the residue sphere model of the Wodak-Janin proce-
dure, made for that softness by blurring the atomic details of the protein structures.