Journal of Molecular Biology
RosettaLigand Docking with Full Ligand and Receptor Flexibility
Introduction
Many biochemical processes depend on the specific interactions between protein receptors and their small-molecule ligands—metabolites, cofactors, hormones, and drugs. Knowledge of the atomic interactions involved is useful both for understanding natural processes and for engineering new therapeutics, but experimentally determining structures of complexes can be difficult, time-consuming, and expensive. Thus, a variety of programs have been developed to computationally “dock” receptors with ligands at varying levels of detail.
Docking programs often ignore conformational changes in the protein receptor, although these are frequently observed upon binding. Better modeling of receptor flexibility is one of the major challenges for the field.1, 2 Of methods that do account for receptor flexibility, most require significant prior knowledge of the system. For instance, 21 different crystal structures were used to generate composite energy grids for docking into HIV-1 protease with AutoDock.3 Similarly, FlexE can combine parts from multiple structures during docking.4 On the other hand, FLIPDock allows many kinds of flexibility to be modeled de novo, but each degree of freedom must be selected manually.5 Likewise, the Glide/Prime method needs only one starting structure, but flexibility is initially modeled simply by truncating three or fewer selected side chains to alanine.6 Such methods are least useful in situations where docking could be most informative: where binding involves interactions and conformations that have not been observed previously and thus could not have been easily foreseen.
Many approaches to docking with receptor flexibility have been recently reviewed by Cozzini and colleagues,7 with a focus on either coarse-grained or atomistic molecular dynamics calculations to generate ensembles of receptor structures. Separate docking calculations can be carried out on each member of the ensemble, or the receptor–ligand energy landscapes can be combined into an ensemble average. Neither approach is perfect: in separate docking, the individual structures constitute only a tiny fraction of all low-energy receptor conformations (even if the algorithm recombines the structures on the fly), while averaging may lead to ligand poses that are not compatible with any real receptor state.7 Both approaches also implicitly assume the ligand-bound conformation is sufficiently low energy in the absence of the ligand to be included in the ensemble. Nonetheless, these approaches have improved performance in some cases.3, 8
Here we describe a new approach to protein–ligand docking that explicitly models full side-chain, backbone, and ligand flexibility with ligand and receptor degrees of freedom explored simultaneously. The new method is an extension of the RosettaLigand algorithm, which uses Monte Carlo sampling and the Rosetta full-atom energy function. We analyze the performance of the new method on an extensive set of retrospective and prospective cross-docking scenarios and find it competitive with and possibly superior to the best existing methods.
Section snippets
Results
As described in Materials and Methods, we have made important enhancements to the RosettaLigand docking program, notably the inclusion of full ligand flexibility (conformers plus torsional minimization), better initial ligand placement, and receptor backbone flexibility. Here we demonstrate, with both retrospective and prospective tests, that these features improve the accuracy of RosettaLigand docking predictions. The retrospective assessment used the Meiler and Baker set: 10 pairs of
Discussion
Overall, we find that the top-scoring ligand pose is correctly positioned within 2 Å RMSD for 64% (54/85) of cross-docking scenarios. To put this in context, Warren et al. studied 19 docking protocols on a diverse set of cross-docking scenarios and found that when considering only the top-scoring pose, even the best program for each receptor rarely placed more than ∼ 60% of compounds within 2 Å RMSD, and on average, each program did significantly worse.12 While it is difficult to meaningfully
Three stages of the docking algorithm
Our docking algorithm is based on the RosettaLigand method described by Meiler and Baker,9 but with enhancements that make it faster and more accurate. It also samples internal degrees of freedom in the ligand and protein backbone, in addition to the rigid-body and protein side-chain degrees of freedom sampled in the original RosettaLigand. The protocol is comprised of three stages, which progress from coarse-grained sampling and scoring to fine optimization and a detailed all-atom energy
Acknowledgements
We thank Andrew Leaver-Fay for design and implementation of core functionality in the new Rosetta code base (“Mini”), Kristian Kaufmann and Jens Meiler for many helpful conversations, and OpenEye Inc. for providing software. I.W.D. gratefully acknowledges support from the University of Washington Genome Training Grant.
References (27)
- et al.
Flexible ligand docking to multiple receptor conformations: a practical alternative
Curr. Opin. Struct. Biol.
(2008) - et al.
Rescoring docking hit lists for model cavity sites: predictions and experimental testing
J. Mol. Biol.
(2008) - et al.
Protein–protein docking with simultaneous optimization of rigid-body displacement and side-chain conformations
J. Mol. Biol.
(2003) - et al.
Protein–protein docking with backbone flexibility
J. Mol. Biol.
(2007) - et al.
Effective energy functions for protein structure prediction
Curr. Opin. Struct. Biol.
(2000) - et al.
An orientation-dependent hydrogen bonding potential improves prediction of specificity and structure for proteins and protein–protein complexes
J. Mol. Biol.
(2003) - et al.
Prediction of protein–ligand interactions. Docking and scoring: successes and gaps
J. Med. Chem.
(2006) - et al.
Protein–ligand docking: current status and future challenges
Proteins
(2006) - et al.
Automated docking to multiple target structures: incorporation of protein mobility and structural water heterogeneity in AutoDock
Proteins
(2002) - et al.
FlexE: efficient molecular docking considering protein structure variations
J. Mol. Biol.
(2001)
FLIPDock: docking flexible ligands into flexible receptors
Proteins
Novel procedure for modeling ligand/receptor induced fit effects
J. Med. Chem.
Target flexibility: an emerging consideration in drug discovery and design
J. Med. Chem.
Cited by (350)
Mechanistic insights into Schizosaccharomyces pombe GT-A family protein Pvg3 in the biosynthesis of pyruvylated β1,3-galactose of N-linked oligosaccharides
2023, Journal of Bioscience and BioengineeringSilent agonists for α7 nicotinic acetylcholine receptors
2023, Pharmacological ResearchCADD, AI and ML in drug discovery: A comprehensive review
2023, European Journal of Pharmaceutical SciencesState-specific protein–ligand complex structure prediction with a multiscale deep generative model
2024, Nature Machine IntelligenceMulti-conformation Aproach of ENM-NMA Dynamic-Based Descriptors for HIV Drug Resistance Prediction
2024, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)