GalaxySite: ligand-binding-site prediction by using molecular docking

Lim Heo 1 Woong-Hee Shin 1 Myeong Sup Lee 0 Chaok Seok 1 0 Department of Biomedical Sciences, University of Ulsan College of Medicine , Seoul 138-736, Korea 1 Department of Chemistry, Seoul National University , Seoul 151-747, Korea Knowledge of ligand-binding sites of proteins provides invaluable information for functional studies, drug design and protein design. Recent progress in ligand-binding-site prediction methods has demonstrated that using information from similar proteins of known structures can improve predictions. The GalaxySite web server, freely accessible at http:// galaxy.seoklab.org/site, combines such information with molecular docking for more precise binding-site prediction for non-metal ligands. According to the recent critical assessments of structure prediction methods held in 2010 and 2012, this server was found to be superior or comparable to other state-of-theart programs in the category of ligand-binding-site prediction. A strong merit of the GalaxySite program is that it provides additional predictions on binding ligands and their binding poses in terms of the optimized 3D coordinates of the protein-ligand complexes, whereas other methods predict only identities of binding-site residues or copy binding geometry from similar proteins. The additional information on the specific binding geometry would be very useful for applications in functional studies and computer-aided drug discovery. - INTRODUCTION Proteins perform their biochemical functions by interacting with other biomolecules such as small ligands, other proteins or nucleic acids. The detection of binding site on a protein makes it possible to infer the function of the protein and provides information on binding pockets crucial for computer-aided drug discovery (1,2). Ligand-binding-site predictions from protein sequences have important implications with regard to sequence-based predictions of the functions of proteins. Binding-site prediction on known experimental protein structures is also important when the known structures do not contain ligands or can bind other ligands. Various evolutionary information-based, geometry-based, energy-based and combined methods have been reported (3). Recently, methods that use experimental structures of similar proteinligand complexes have been successfully applied in binding-site predictions in critical assessment of structure prediction (CASP) experiments (47). In such methods, binding-site information of homologous proteins of known structures is utilized by assuming that similar proteinligand contacts occur in the target protein. These methods predict only ligand-binding residues or ligandbinding geometry based on simple structure superimposition to similar proteinligand complexes (812). In this paper, we introduce a new method that uses such information in the context of proteinligand docking. Because specific binding of ligands to proteins occurs owing to favorable physicochemical interactions, it can be expected that binding-site prediction based on physical chemistry using molecular docking can provide predictions that are more precise. In addition to revealing the identities of the contacting residues, molecular docking can also provide detailed information on atomic interactions between protein and ligand in terms of the optimized 3D coordinates of the proteinligand complex. The binding geometry obtained by docking can be different from the geometry obtained by simple structure superimposition with similar proteins, and the binding pose optimized by docking tends to have physically more realistic geometry with no severe steric clashes. Such precise information would be very useful for the prediction of specific functions and applications in drug discovery. However, a few difficulties have to be overcome to apply molecular docking to binding-site prediction methods. First, docking requires prior knowledge of the protein structure and binding ligand. Second, docking results can be sensitive to structural details, and the prediction accuracy may decrease if the protein structure is not sufficiently accurate or if conformational changes occur upon binding (8). In the GalaxySite program, binding ligand is predicted using a similarity-based method, and the protein structure is provided by the user or predicted from a template*To whom correspondence should be addressed. Tel: +82 2 880 9197; Fax: +82 2 889 1568; Email: Correspondence may also be addressed to Myeong Sup Lee. Tel: +82 2 3010 2979; Email: based modeling method. The current binding-site prediction method is accurate even when only chemically similar ligands are predicted. The energy function for docking is designed to be less sensitive to structural details by adapting a combination of physics-based terms of AutoDock3 (13) and restraint terms derived from homologous protein ligand complexes of known experimental structures. GalaxySite has been tested on the following nonmetal ligand-binding-site prediction test (...truncated)