Study On Protein-Protein Docking With Backbone Flexibility

Computational models of protein-protein clocking that incorporate backbone flex-ibility can predict perturbations of the backbone and side chains during docking and produce protein-protein interaction models with atomic accuracy. The challenges faced when modeling protein-protein interactions include the accuracy of the binding site prediction, the accuracy of the flexible region prediction, the choice of effective sam-pling strategy, and the type of computational model. These choices constitute the greatest challenges for theoretical computation of protein-protein docking. This paper-studies the two hard problems, one is the automatic prediction of flexible regions in protein-protein docking, the other is the docking protocol developed bathed on the characteristics of the flexible regions.First, this paper proposes a general method and the corresponding prediction mechanism to identify the flexible regions that consists of the domain-linked flexible hinges which causes the domain motions and the flexible loops which causes the local backbone conformational changes automatically to serve for the protein-protein dock-ing with backbone flexibility. Second, we develop the docking protocol based on the characteristics of the identified flexible regions and conduct two docking experiments, that is the protocol of docking with loop minimization and docking with domain as-sembly. Finally, we evaluate the performance of the flexible regions identified and the performance of the docking protocols developed by us based on the docking results.The protocol of docking with loop minimization at the high-resolution stag is evaluated based on a test set which is derived directly from the renowned journal internationally. Comparison of results on this test set between algorithms with and without prediction of flexible regions demonstrate the superior recover of energy fun-nels in many target interactions using the new loop refinement model based on our docking protocol. Our decoys are superior for each target with the total number of satisfactory models almost double that of other programs, these results demonstrate that the quality of our predictions is acceptable and superior to those of the competitor whether energy funnels are achieved or not and the models selected whether by inter-face energy or by combined, energy proposed by us also demonstrate the high quality of the decoy sets as well as the docking models generated this paper.The protocol of docking with domain assembly is evaluated based on a test set consisting of four targets and each of these four targets is classified as the difficulty case internationally. The protocol of docking with domain assembly by fragment insertion in the identified flexible hinges at the low-resolution stage and using loop refinement upon the docking with loop minimization in the identified flexible loops at the high-resolution stage on these four hard targets this paper also show encouraging results one target exhibits energy funnel for the results selected whether by interface energy or combined energy, the other two both meet the conditions of acceptable accuracy, and the last which is the most difficult of these four examined, the criterion of Fnat of the best model in the decoy set meets the conditions of acceptable accuracy but for Lrmsd and Irmsd which are only very close to it.All these docking results this paper demonstrate that the automatic prediction of flexible backbone regions can greatly improve the performance of protein-protein docking models, and the docking protocols developed by us based on the characteristics of the flexible regions are correct and feasible.