Flexible Relaxation of Rigid-Body Docking Solutions Marcin Kro ´ l, 1,2* Alexander L. Tournier, 1 and Paul A. Bates 1 * 1 Biomolecular Modelling Laboratory, Cancer Research UK London Research Institute Lincoln’s Inn Fields Laboratories, 44 Lincoln’s Inn Fields, London WC2A 3PX, United Kingdom 2 Department of Bioinformatics and Telemedicine, Collegium Medicum, Jagiellonian University, Krako ´w 31-501, Poland ABSTRACT Molecular Dynamics (MD) simula- tions have been performed on a set of rigid-body docking poses, carried out over 25 protein–protein complexes. The results show that fully flexible relaxation increases the fraction of native contacts (NC) by up to 70% for certain docking poses. The largest increase in the fraction of NC is observed for docking poses where anchor residues are able to sample their bound conformation. For each MD simulation, structural snap-shots were clustered and the centre of each cluster used as the MD- relaxed docking pose. A comparison between two energy-based scoring schemes, the first calculated for the MD-relaxed poses, the second for energy minimized poses, shows that the former are better in ranking complexes with large hydrophobic interfaces. Furthermore, complexes with large interfaces are generally ranked well, regardless of the type of relaxation method chosen, whereas complexes with small hydrophobic interfaces remain difficult to rank. In general, the results indicate that current force-fields are able to cor- rectly describe direct intermolecular interactions between receptor and ligand molecules. However, these force-fields still fail in cases where protein– protein complexes are stabilized by subtle energy contributions. Proteins 2007;68:159–169. V V C 2007 Wiley-Liss, Inc. Key words: protein interactions; flexible docking; molecular dynamics; scoring funtions; induced fit INTRODUCTION Protein–protein interactions play a crucial role in many biological processes such as signalling, regulation, and immunogenic recognition. 1 Identification of these interactions can be performed by both experimental and theoretical techniques. However, while there is a large amount of experimental information on potential pro- tein–protein interactions at the proteome scale, 2,3 only a small fraction of protein–protein complexes have been investigated at the atomic level. Atomic level detail is needed to fully understand the functional subunits involved in protein–protein interactions. Therefore, a number of computational approaches have been devel- oped to predict the structure of a protein complex, given the structures of the unbound components. 4,5 These docking methods are thoroughly tested in the commu- nity-wide blind docking experiment CAPRI. 6 As indi- cated by the results of CAPRI, computational methods are relatively successful in predicting complex structures in cases where conformational change upon binding is small (less than 1.0 A ˚ backbone RMSD) using rigid-body techniques. However, the majority of these approaches are much less successful if proteins change their confor- mations upon complex formation (the induced-fit effect). Therefore, much effort is now invested in taking account of the protein flexibility in docking. Docking methods currently used include soft-core docking 7 ; ensemble docking where many different confor- mations of the unbound receptor and ligand are gener- ated and subsequently cross-docked 8,9 ; and explicit treat- ment of sidechain and, less frequently, backbone flexibil- ity. 10–12 Promising methods have also been reported using a multicopy approach to model loop rearrange- ments 13 and global deformations along the precalculated soft, collective degrees of freedom. 14 For a good review of the subject the reader is referred to Bonvin 2006. 15 In our earlier work we implicitly included protein flex- ibility by generating ensembles of structures for each binding protein. 8 Ensembles were generated by running explicit solvent (ES) molecular dynamics (MD) simula- tions. Subsequently, generated sets of structures were used for cross-docking. We have shown that when start- ing from the unbound state MD simulations did not sam- ple the bound state entirely. However, the inclusion of ensembles of the receptor and ligand in the rigid-body docking was partially successful when compared with docking the unbound conformations. The fact that none of the studied proteins sampled the complete bound state is expected since the ensemble of structures was gener- ated without the presence of the other protein and, therefore, the induced-fit effect was neglected. Conse- The Supplementary Material referred to in this article can be found at http://www.interscience.wiley.com/jpages/0887-3585/suppmat *Correspondence to: Marcin Kro ´l, Biomolecular Modelling Labora- tory, Cancer Research UK London Research Institute Lincoln’s Inn Fields Laboratories, 44 Lincoln’s Inn Fields, London WC2A 3PX, UK. E-mail: mykrol@cyf-kr.edu.pl or Paul A. Bates, Biomolecular Model- ling Laboratory, Cancer Research UK London Research Institute Lin- coln’s Inn Fields Laboratories, 44 Lincoln’s Inn Fields, London WC2A 3PX, UK. E-mail: paul.bates@cancer.org.uk Received 6 September 2006; Revised 23 November 2006; Accepted 24 December 2006 Published online 30 March 2007 in Wiley InterScience (www. interscience.wiley.com). DOI: 10.1002/prot.21391 V V C 2007 WILEY-LISS, INC. PROTEINS: Structure, Function, and Bioinformatics 68:159–169 (2007)