Journal of Molecular Biology
Volume 385, Issue 2, 16 January 2009, Pages 381-392
Journal home page for Journal of Molecular Biology

RosettaLigand Docking with Full Ligand and Receptor Flexibility

https://doi.org/10.1016/j.jmb.2008.11.010Get rights and content

Summary

Computational docking of small-molecule ligands into protein receptors is an important tool for modern drug discovery. Although conformational adjustments are frequently observed between the free and ligand-bound states, the conformational flexibility of the protein is typically ignored in protein–small molecule docking programs. We previously described the program RosettaLigand, which leverages the Rosetta energy function and side-chain repacking algorithm to account for flexibility of all side chains in the binding site. Here we present extensions to RosettaLigand that incorporate full ligand flexibility as well as receptor backbone flexibility. Including receptor backbone flexibility is found to produce more correct docked complexes and to lower the average RMSD of the best-scoring docked poses relative to the rigid-backbone results. On a challenging set of retrospective and prospective cross-docking tests, we find that the top-scoring ligand pose is correctly positioned within 2 Å RMSD for 64% (54/85) of cases overall.

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)

  • ZhaoY. et al.

    FLIPDock: docking flexible ligands into flexible receptors

    Proteins

    (2007)
  • ShermanW. et al.

    Novel procedure for modeling ligand/receptor induced fit effects

    J. Med. Chem.

    (2006)
  • CozziniP. et al.

    Target flexibility: an emerging consideration in drug discovery and design

    J. Med. Chem.

    (2008)
  • Cited by (350)

    • CADD, AI and ML in drug discovery: A comprehensive review

      2023, European Journal of Pharmaceutical Sciences
    • Multi-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)
    View all citing articles on Scopus
    View full text