Sampling low-energy protein-protein configurations with basin hopping

Here we propose a novel algorithm to efficiently generate near-native configurations of dimers. The algorithm addresses rigid protein-protein docking as an optimization problem and build upon the Basin Hopping framework to sample bound configurations that correspond to low energy local minima in the dimeric energy surface. At its core, the algorithm is driven by a geometric treatment. The unbound structures are each analyzed to represent their surfaces through a sparse set of critical points. Triangles are defined over these points to identify regions of geometric complementarity between the two monomeric structures. Alignment of two geometrically-complementary triangles results in a rigid-body transformation that bounds the two monomers to each-other, an efficient process referred to as geometric hashing. The set of rigid-body transformations can be reduced by focusing only on those that align active triangles of evolutionary-conserved critical points.