Protein­protein docking algorithms aim to predict how two proteins interact with each other to form a complex. Docking algorithms need to fulfil two main tasks: (1) sampling the possible relative positions of the two proteins and (2) computing the interaction energy at each position. Although the docking problem has been studied for over 25 years, doing this accurately and thoroughly remains a computationally expensive task. We are therefore developing a new algorithm to face the protein docking problem in a novel and efficient way. We are interested in performing an exhaustive search of the rigid­body docking space, where the positioning of the two proteins is not driven by any prior knowledge (e.g. from homology), while still using an accurate force field interaction energy model. In order to reduce the O(N^2) cost of atomistic force­field models, we use a coarse­grained (CG) bead model taken from ATTRACT [1]. However, a naive energy calculation for every trial orientation would still cost O(N^2) energy evaluations in the number of beads. We are therefore developing a method to detect the locations of all possible clashes before performing any energy calculations, thus allowing us to avoid calculating energies for many millions of useless trial orientations. Based on a preliminary study of protein­protein interfaces in complexes from the Protein Docking Benchmark [2], we found that a large number of interfaces contain at least one pair of CG beads at almost their optimal distance. Therefore, our idea is to perform a series of restricted docking searches in which one surface bead from each docking partner is placed in contact at the coordinate origin. This leaves a 3D rotational search, in which ligand may rotate around a fixed receptor. Of course, a full docking search requires all possible pairs of surface beads to be placed together. However, within each rotational sub­problem, we can exploit the fact that rotations do not change any distances from the origin. Thus, we can pre­calculate a "3D rotational map" of all of the rotations that will cause the beads to clash. We can then restrict the remaining rotational search and energy calculation to a small region near the forbidden rotations in the clash map.