Protein-Protein Docking with Simultaneous Optimization of Rigid-body Displacement and Side-chain ConformationsIntroductionResultsDocking algorithmPerturbation studiesGlobal searches using unbound componentsDiscussionConclusionMaterials and MethodsBenchmark setLow-resolution scoring functionsResidue environment and residue pair potentialsHigh-resolution scoring functions and refinement detailsAcknowledgementsReferencesProtein–Protein Docking with SimultaneousOptimization of Rigid-body Displacement andSide-chain ConformationsJeffrey J. Gray, Stewart Moughon, Chu Wang, Ora Schueler-FurmanBrian Kuhlman, Carol A. Rohl and David Baker*Howard Hughes MedicalInstitute and Department ofBiochemistry, University ofWashington, J-567 HealthSciences, Box 357350, SeattleWA 98195, USAProtein–protein docking algorithms provide a means to elucidate struc-tural details for presently unknown complexes. Here, we present andevaluate a new method to predict protein–protein complexes from thecoordinates of the unbound monomer components. The method employsa low-resolution, rigid-body, Monte Carlo search followed by simul-taneous optimization of backbone displacement and side-chain confor-mations using Monte Carlo minimization. Up to 105independentsimulations are carried out, and the resulting “decoys” are ranked usingan energy function dominated by van der Waals interactions, an implicitsolvation model, and an orientation-dependent hydrogen bonding poten-tial. Top-ranking decoys are clustered to select the final predictions.Small-perturbation studies reveal the formation of binding funnels in 42of 54 cases using coordinates derived from the bound complexes and in32 of 54 cases using independently determined coordinates of one orboth monomers. Experimental binding affinities correlate with the calcu-lated score function and explain the predictive success or failure of manytargets. Global searches using one or both unbound components predictat least 25% of the native residue–residue contacts in 28 of the 32 caseswhere binding funnels exist. The results suggest that the method maysoon be useful for generating models of biologically important complexesfrom the structures of the isolated components, but they also highlight thechallenges that must be met to achieve consistent and accurate predictionof protein–protein interactions.q 2003 Elsevier Ltd. All rights reservedKeywords: protein–protein docking; protein binding; biomolecularmodeling; biomolecular free energy functions; conformational change*Corresponding authorIntroductionThe protein docking problem, that is, the task ofassembling two separate protein components intotheir biologically relevant complex structure, isimportant for several reasons. First, it is of extremerelevance to cellular biology, where function isaccomplished by proteins interacting with them-selves and with other molecular components.Second, the protein docking problem presents afundamental test of our understanding of theenergetics of macromolecular interactions, as thenative complex structure is almost certainly at aglobal free energy minimum. Finally, an importantpost-genomic goal is the characterization of thestructures of protein–protein complexes, and com-putational tools offer an inexpensive means tocarry out large-scale studies.Protein–protein docking has been studied forsome time now, and there are several excellentreview articles available.1–3Many early and currentdocking strategies involve grid-based searchalgorithms.4–11These algorithms are quite success-ful at joining the components of a separatedcomplex because of the excellent shape comple-mentarity at the interface. However, proteins andprotein interfaces are flexible, and the confor-mations of the bound partners often differ from0022-2836/$ - see front matter q 2003 Elsevier Ltd. All rights reservedPresent address: J. J. Gray, Chemical and BiomolecularEngineering, Johns Hopkins University, 3400 N. CharlesSt., Baltimore, MD 21218, USA.E-mail address of the corresponding author:[email protected] used: rmsd, root-mean-squareddistance.doi:10.1016/S0022-2836(03)00670-3 J. Mol. Biol. (2003) 331, 281–299those of the isolated components. If the unboundmonomer components are used, it is no longertrivial to match the shapes together. Strategies toaddress this include softening the interface orcoarsening the grid to allow more uncertainty inthe matching process.8,12,13Chemical and physicalinformation can be incorporated by including thisinformation while matching the surfaces.6,10,11,14Finally, some algorithms explicitly include side-chain flexibility, although in most cases in onlyone of the protein partners.15 – 17Accurate and con-sistent prediction of correct complex structuresfrom unbound components remains elusive, andfew algorithms have been tested on large sets oftargets.8,14,17Recent large-scale studies have examined up to27 targets. Ferna´ndez-Recio et al.17tested theirmethod on unbound components using soft dock-ing with side-chain refinement, localizing thesearch space to one-half of the receptor. Theiralgorithm admirably found correct solutions in thetop 20 models in 17 of 24 cases, including seven of11 protease-inhibitor cases for which the top-ranked solution was correct. Palma et al.8similarlyfound correct solutions of rank 20 or less in 14 of25 cases (bound, semi-bound and unbound) usinga soft docking algorithm designed to capture side-chain flexibility. Chen & Weng14used target func-tions that are tolerant of conformational change tostudy 27 systems; they predicted 12 structureswithin the top 20 ranked decoys, and three systemsfor which the correct solution was top-ranked.While these results are encouraging, current searchalgorithms are not sufficient to efficiently exploreconformational space, and free energy functionsare unable to consistently recognize correct com-plexes. There are still unsolved problems in thefield of protein–protein docking, and insightcould come from new approaches.Like protein docking, protein folding requires avast search and an accurate free energy or scoringfunction. Recently, progress has been noted insingle-protein, ab initio structure predictionalgorithms.18In particular, the Rosetta programdeveloped at the University of Washington is nowable to construct crude (, 5A˚) models of manyshort (less than 150 amino acid residues) proteinsequences.19This progress has emerged throughthe application of the following core philosophiesand
View Full Document
Unlocking...