A multistage approach to protein-protein docking

点击翻译按钮获取中文摘要

Protein-protein interactions are integral to virtually all biological pathways. Many important interactions occur in weak, transient complexes that will not be amenable to direct experimental analysis, even when both proteins can be isolated and their structures determined. Thus, it is important to develop computational docking methods which, starting from the structures of component proteins, can determine the structure of their complexes. We have developed a multistage docking algorithm that provided the best results in the latest rounds of the CAPRI (Critical Assessment of Predicted Interactions) worldwide docking experiment. In addition, our docking server ClusPro was the best among automated servers. Although the CAPRI results demonstrate progress, a number of major problems remain unsolved. First, docking homology models is a challenge and all methods used in CAPRI performed poorly for such targets. Docking unbound structures is also difficult if binding is accompanied by substantial backbone conformational change. Second, it is not clear whether a model generated by docking represents a specific and stable complex. Third, the interface may include regions that are disordered in the separate proteins, challenging docking methods. We address these problems by pursuing three specific aims. First, we develop a novel algorithm for docking homology models and proteins with substantial backbone flexibility. The method is based on the hypothesis that the interface in complexes is sequentially and structurally more conserved than the rest of the proteins. Since such regions are frequently sufficient for recognition, identification and correct docking of the key segments can yield near-native docked structures. For homology models this implies that one can dock the regions that can be reliably modeled, and then expand the models by adding back the removed parts using the docked structures as constraints. The problem of docking "difficult cases" with substantial backbone conformational change can also be addressed by identifying and docking the structurally most conserved regions. Once clusters of the docked rigid fragments are obtained, the models are expanded by rebuilding the more flexible parts. Second, we use a two-step approach to examine the stability of protein complexes, first by removing small and hence unlikely clusters of low energy docked structures, and then by calculating dissociation rates by stochastic roadmap simulation. The method will be validated on a benchmark set that includes models of real protein complexes and decoys generated by docking non-interacting protein pairs. The approach will also be used to determine whether complex structures deposited to the PDB are biologically relevant. Third, we consider the problem of determining the structure of flexible loops and/or disordered regions when they become parts of a protein- protein interface. Rather than attempting to predict and to dock the most likely conformation of the flexible fragment, we build their bound structure directly into binding hot spots of the partner protein. Flexible peptide docking methods will be used to expand the docked fragments by adding further residues.