Constraining Redundancy to Improve Protein Docking

Predicting protein-protein complexes (protein docking) is an important factor for understanding the majority of biochemical processes. In general, protein docking algorithms search through a large number of possible relative placements of the interacting partners, filtering out the majority of the candidates in order to produce a manageable set of candidates that can be examined in greater detail. This is a six-dimensional search through three rotational degrees of freedom and three translational degrees of freedom of one partner (the probe) relative to the other (the target). The standard approach is to use a fixed step both for the rotation (typically 10\(^\circ \) to 15\(^\circ \)) and the translation (typically 1A). Since proteins are not isotropic, a homogeneous rotational sampling can result in redundancies or excessive displacement of important atoms. A similar problem occurs in the translational sampling, since the small step necessary to find the optimal fit between the two molecules results in structures that differ by so little that they become redundant. In this paper we propose a constraint-based approach that improves the search by eliminating these redundancies and adapting the sampling to the size and shape of the proteins involved. A test on 217 protein complexes from the protein-protein Docking Benchmark Version 5 shows an increase of over 50 % in the average number of non-degenerate acceptable models retained for the most difficult cases. Furthermore, for about 75 % of the complexes in the benchmark, computation time is decreased by half, on average.