A Fragment And Graph Theory Based Ligand-Protein Docking Approach And Docking Tests With Homology Model Receptor Structures

Computational molecular docking is extensively used in studying protein-ligand binding and in structure-based drug design.In protein-ligand docking,the three-dimensional structure of the receptor protein is of key importance.Common docking protocols perform the best when they are applied to a crystal structure of the receptor determined in complex with a ligand that is highly similar to the target ligand.In some benchmarking studies,the input receptor structures have been determined in complex with the respective target ligands themselves.These calculations are referred to as bound docking.Real-world applications do not correspond to bound docking,because an experimental structure of the target ligand-receptor complex would be one of the main result to seek for,not available as an input.Such a calculation corresponds to unbound docking in which the available receptor structure may be that of an apo protein or from a complex of the receptor bound to a different ligand.In more extreme situations of unbound docking,no experimental structure of the receptor is available and docking needs to be performed on receptor structures constructed through homology modelling.Since modelled structures are usually imperfect and cannot match high-resolution crystal structures in accuracy,the performance of common docking algorithms may degenerate significantly when they are applied to modelled receptor structures,especially for docking larger ligands.In the present work,we describe a fragment-based docking method that is less sensitive to the accuracy of the receptor structures than commonly available docking approaches.In this method,molecular fragments instead of complete ligands are docked to a receptor structure first.The fragments are chosen to be small and rigid,so that they may explore the binding pockets more efficiently and thoroughly compared with complete ligands.As different fragments explore their favorable binding sites independently,the results may be less sensitive to the overall accuracy of a modelled receptor structure.After obtaining a set of high-scoring docked poses of the fragments,we apply a graph-theory based approach to find good structural alignments between a complete ligand and the binding poses of its composing fragments.This leads to candidate poses for the complete ligand.In these poses the fragments composing the ligand are located close to energetically favorable binding sites that have been identified through fragment docking.The resulting poses may be further optimized,scored and ranked.To establish the new docking method,we first tested if fragment docking with AutoDock can provide docked poses that well cover actual binding locations of respective fragments in known ligand-receptor complexes.Twenty fragments have been considered for this purpose.The success rate for the docked fragment poses to reproduce those in known ligand-receptor complexes was 75%when considering the top 20%of results based on AutoDock scores.Then we benchmarked the complete docking process by considering a number of CYP450-substrate complexes.One set of benchmarking was performed for bound docking,namely,on CYP450 structures determined in complex with the respective target substrates.The other set was carried out on modelled structures of CYP450 proteins.Results were compared with those obtained by applying Autodock on complete ligands.It was found that our method successfully yielded experimental result-like binding poses for all systems.The final structures were ranked using binding scoring functions and the success rates judged by criteria of retaining a varied number of highest-scoring results were determined.In the case of bound docking,the success rate was 100%when top 3%hits were analyzed.For unbound docking with modelled structures,the success rates were 50%and 25%when 20%and 10%of top ranked structures were analyzed,respectively.We also found our method is particularly advantageous in terms of docking larger ligands into modelled structures.For one such ligand,the minimum RMSD of 0.84 A from the actual ligand pose was obtained with the fragment-based approach in comparison with a minimum RMSD of 5.68 A obtained using AutoDock.Since a ligand can be present in different conformations and we may not know which one is the bound conformation,we also applied our method on different ligand conformations for one of the examples to see if we can recover the experimentally determined binding pose.The result was encouraging with the minimum RMSD docked pose deviated by an RMSD of 1.37 A from the actual ligand pose.Our results suggest that the fragment and graph-theory based approach introduced here can be an effective method for docking large ligands into protein receptor structures constructed through homology modelling.