Macromolecular Modeling Blog ™

This is the second post in the CAPRI series, summarizing the presentations of Sandor Vajda, Alexandre Bonvin, and Julie Mitchell, as provided by the speakers. More to appear in the continuation of the series.

Predictions with PIPER, SDU and stability analysis at CAPRI and our server ClusPro.

D. Kozakov, D. Hall, R. Brenke, D. Beglov, Y. Shen, J. Zhang, K. Li, S. Comeau, S. Vajda

Since the last CAPRI evaluation meeting, our group has made advances in both our initial sampling of conformations and the refinement methods used for choosing models.  Our ClusPro server has moved from using DOT and ZDOCK in version 1 to our own FFT-based docking program, PIPER, in version 2. This new server uses a similar clustering method, but PIPER introduces several innovative features such as a DARS potential, allowing us to keep only 1000 conformations for clustering as opposed to the 2000 previously retained when using ZDOCK. The server also has special modes for docking antibody-antigen complexes and basic support for placing attractions and repulsion on individual residues. Following sampling with PIPER and clustering, we explore the energy landscape around the cluster centers to help in selecting, ranking, and refining models.  We have made improvements to refinement using SDU, and, in conjunction with Ora Schueler-Furman, developed a new stability analysis algorithm using Rosetta.

During the rounds since the last evaluation, our group and server have performed extremely well. We were the only group to have high or medium predictions for six of the targets, while ClusPro v.2 was the only server to predict five targets.  When looking at the overall CAPRI results, there were six interfaces on five targets (T32, T34,T40.CA, T40.CB, T41, T42_AB) with 10 or more medium or high predictions.  Our group was able to give medium or high predictions for all of these, ranked as the top model for five, while ClusPro predicted five of these interfaces.  For T32, where ClusPro did not produce an acceptable model, its largest cluster was still in the correct location and was refined to our group’s high quality prediction.  Thus, the latest rounds of CAPRI show our methods generally work for the widely “predictable” interfaces.

On T39, our group and server provided 2 of the 3 correct predictions among all CAPRI participants.  This target was interesting as there was misleading experimental information indicating binding should occur to the GAP domain of Centaurin-alpha 1. This information was not used in docking with ClusPro, which gave a medium quality prediction as the tenth ranked cluster.  When the stability analysis was applied to this model, it was found to be “ultrastable”, showing more stability than anything in the Protein-Protein Docking Benchmark, but the experimental information caused us to rank it 7th.  This target gives some argument for trusting our methods despite some contradictory evidence.  Too often, computations want to treat experimental information as truth, rather than weighting it, with some variation based on confidence in the method and the person carrying out the experiments.

S.J. de Vries, A.S.J. Melquiond, M. van Dijk, A.M.J.J. Bonvin

HADDOCK [1-3] is a popular program for data-driven docking and modeling of biomolecular complexes, which supports a wide-range of experimental data. The CAPRI presentation of the HADDOCK group was mainly focused on the strengths and weaknesses of our program, explained in the context of the CAPRI challenge. In particular, it was demonstrated that the use of experimental data to drive the docking is a very powerful tool (as, for example, in target 40, for which HADDOCK obtained 10 three-star solutions and identified correctly the two binding sites), but can also have drawbacks (as was the case in target 32, for which no correct results were generating, because the bioinformatics predictions covered the interface defined by the two inhibitors while only one was considered for the docking based on the provided information). However with 6 successful predictions HADDOCK is performing among the top groups.
The HADDOCK web-server was also presented (available through the HADDOCK portal) and its multi-body docking implementation, especially developed for CAPRI target 42. The server has both a simple, user-friendly interface that requires only the structures of the individual components and a list of interacting residues as an input and can be used by “docking newbies” and more complex, versatile interfaces giving access to every single parameter used during the docking, suitable for very expert dockers. The server has participated in CAPRI rounds 15-19 making four acceptable predictions: one star for T35 (protein-RNA) and T42, two-stars for T41 and three-star for T40.

Finally, the question was addressed whether the scoring functions used in docking are ready to predict interactomes. Various scoring functions (ZDOCK, Rosetta, HADDOCK, PISA, DFIRE, FastContact, FireDock) were tested on a new binding affinity benchmark consisting of 84 complexes. All showed poor correlations (R2 ~ 0) indicating that the prediction of binding affinity is very much still an open challenge.

Dominguez, C., Boelens, R., & Bonvin, A. (2003). HADDOCK: A Protein?Protein Docking Approach Based on Biochemical or Biophysical Information Journal of the American Chemical Society, 125 (7), 1731-1737 DOI: 10.1021/ja026939x
van Dijk, A., de Vries, S., Dominguez, C., Chen, H., Zhou, H., & Bonvin, A. (2005). Data-driven docking: HADDOCK’s adventures in CAPRI Proteins: Structure, Function, and Bioinformatics, 60 (2), 232-238 DOI: 10.1002/prot.20563
de Vries, S., van Dijk, A., Krzeminski, M., van Dijk, M., Thureau, A., Hsu, V., Wassenaar, T., & Bonvin, A. (2007). HADDOCK versus HADDOCK: New features and performance of HADDOCK2.0 on the CAPRI targets Proteins: Structure, Function, and Bioinformatics, 69 (4), 726-733 DOI: 10.1002/prot.21723

ReplicOpter: A Replica Optimizer for Flexible Docking

O.N.A Demerdash, A. Buyan and J.C. Mitchell

Julie Mitchell presented a new method, ReplicOpter, that her group has developed for flexible refinement of rigid docking predictions.  ReplicOpter is computationally efficient and reflects observed motions within a protein’s structural class. Using structural homologs of a given protein, a family of deformation models that capture likely motions is derived. A small number of models can generate a much larger number of conformers, by exchanging each flexible region independently of the others.

ReplicOpter can perform hierarchical clustering from a list of rigid docking predictions and find nearby structures to any promising cluster representatives. These predicted complexes can then be optimized and rescored in order to attain good predictions. ReplicOpter’s scoring/energy function includes a Lennard-Jones potential that is softened using the Anderson-Chandler-Weeks decomposition, a desolvation term derived from the Atomic Contact Energy function, Coulombic electrostatics, a hydrogen bonding energy, and terms to indicate the presence of pi-pi and pi-cation interactions. ReplicOpter has performed well on several recent CAPRI systems.

---

Read also the previous post in the series: CAPRI or: What is the State of Protein-Protein Docking?