(626a) Flexible-Backbone Protein Docking Using Motif Scoring and Large Conformational Ensembles

Computational prediction of protein-protein complex structures facilitates a fundamental understanding of biological mechanisms and enables drug design. Binding-induced conformational changes in proteins confound current protein-protein docking algorithms by greatly increasing the degrees of freedom to be sampled. While rotamer libraries have alleviated the sampling challenges for side-chains, backbone flexibility remains a principal challenge in docking. Previous studies have found limited success by varying the backbone along a restricted set of coordinates or residues or by docking a small number of backbones conformations of the two partners. In this study, we develop and benchmark the new EnsembleDock application in the Rosetta Molecular Modeling Suite. During the first, coarse-grained step, the method explores the relative orientation between the monomers while rapidly sampling from hundreds of backbone conformations of the unbound monomers, pre-generated using a variety of techniques to capture different backbone motions. It adapts to the backbone variance in the ensembles to ensure adequate backbone sampling for diverse ensembles; up to O(10⁶) combinations of backbone conformations and relative orientations are tested in 1-4 minutes. We have improved the near-native discrimination for these conformations by as much as 75% over the previous version by using motif scoring, a scheme based on a pre-tabulated set of implicit side-chain energies. In the second, all atom step, the interface is optimized by packing the side chains while performing small relative motions between the two monomers. To more aggressively simulate induced changes at the interface, we have developed an optional step where all atoms at the interface are moved along energy gradients in a reduced-coordinate framework. With these new methods, EnsembleDock can capture interface backbone changes greater than 1 Å between the unbound and bound states without any geometric restrictions. For targets with backbone changes greater than 2 Å, we verify the ability of the method to pick out near-native backbones by doping the existing ensembles with near-bound backbones.