CAPRI: Selected Talks II

Jan
17
2010

This is the third post in the CAPRI series, summarizing the presentations of Jeffrey Gray, Zhiping Weng, and Miriam Eisenstein, as provided by the speakers. More to appear in the continuation of the series.

Towards Accurate Homology Model Docking with RosettaDock, EnsembleDock & SnugDock in CAPRI Rounds 13-19

Sidhartha Chaudhury, Krishna Praneeth Kilambi, Aroop Sircar & Jeffrey J. Gray

While protein-protein docking has enjoyed considerable success, backbone flexibility and uncertainty in homology models remain significant challenges. In this talk, we will summarize our recent CAPRI predictions and the associated new RosettaDock approaches that we have developed both for CAPRI and for general application. Overall we had 5 successful predictions including 3 high quality predictions 1 medium quality and an acceptable protein-RNA prediction.

Difficulty in docking homology models underscore the need to address backbone flexibility not only to capture binding induced conformational changes but also to ensure robustness to uncertainties in starting structures. We have recently developed techniques to dock ensembles of backbones simultaneously and found this approach to be effective on a benchmark of NMR docking targets. This technique proved useful for one of the targets where we generated a number of predictions that closely match a homologous complex structure.

Ensemble of conformers during the docking procedure.

We have also developed a technique called SnugDock which is tailored to antibodies and combines multi-body docking, loop construction and minimization to optimize the paratop. Benchmark tests encouragingly show that SnugDock combined with EnsembleDock can dock homology antibodies with results comparable to docking of crystal structures.

Although SungDock is designed for antibodies we tailored the flexible loop methods for some of the targets. In target 32 SnugDock improved the monomer backbone towards the bound conformation to achieve a high quality prediction with the best Ligand RMSD and Interface RMSD of all submitted predictions. Other methodological developments included adapting docking for RNA (3 acceptable predictions for target 34) exploiting C3 symmetry in homotrimer global docking and incorporating experimental data.

Failure analysis shows that for several targets local energy funnels are observed post-hoc but global searches failed to find near-native structures. RosettaDock does very well for local searches. Further analysis shows that using an interface based score vs. a total score, is better when backbone flexibility is involved.

Chaudhury S, & Gray JJ (2008). Conformer selection and induced fit in flexible backbone protein-protein docking using computational and NMR ensembles. Journal of molecular biology, 381 (4), 1068-87 PMID: 18640688
*SnugDock to appear in PLoS Computational Biology.

Performance of ZDOCK in CAPRI Rounds 13 – 19

Howook Hwang, Thom Vreven, Brian Pierce, Jui-Hung Hung, Zhiping Weng

In the 4th CAPRI meeting, we report on the performance of ZDOCK3.0 and ZRANK in CAPRI rounds 13-19, and introduce the Atom Contact Frequency (ACF) analysis based on ZDOCK3.0 predictions. The ACF analysis allows one to predict the interface areas of the complexes, and in combination with biological data available in the literature, proved a valuable addition to the docking pipeline. Furthermore, we incorporated a straightforward and efficient clustering algorithm to remove redundancies from the final set predictions. With the combined application of these methods, we achieved at least one acceptable prediction for six targets (two acceptable, two medium quality and two high quality predictions) out of 12 targets.

Links:

Hwang H, Pierce B, Mintseris J, Janin J, & Weng Z (2008). Protein-protein docking benchmark version 3.0. Proteins, 73 (3), 705-9 PMID: 18491384
Pierce B, & Weng Z (2008). A combination of rescoring and refinement significantly improves protein docking performance. Proteins, 72 (1), 270-9 PMID: 18214977

Anchoring spots mapping and post-scan filtering: New tools for better detection of correct docking models.

Avraham Ben-Shimon, Noga Kowalsman, Miriam Eisenstein

Background: MolFit is a grid based FFT approach, which incorporate into the correlation function an electrostatic term as well as an hydrophobic complementarity term, these can be weighted according to external information. The combined score is used to rank the models.

Recently we developed several tools that significantly improve our ability to detect correct and nearly correct docking models. this include a post scan filtering and re-evaluation procedure and an anchoring spots mapping algorithm.

Descriptors as filters for false positive docking solutions.

Our class specific post scan processing dramatically enhances the discrimination between nearly correct and false docking models. it employs several propensity descriptors calculated for the interface core and interface core clusters count. These descriptors (Residue propensity: the preference of a residue to appear in the interface. Residue-residue propensity: the preference of two interface residues to interact with each other. Solvation energy: the energy gained or lost upon association of molecules of solvent with molecules of solute.) which add to the in-scan geometric-electrostatics-hydrophobic complementary score are used in two ways: as classifiers in a “soft intersection filter” that eliminates most of the false models and in a new scoring function that underscores the correct model. Two important points are that each descriptor captures different false positives, ergo their combination is a powerful filter (see image) and that for different classes of interactions (e.g. Enzyme-Inhibitors, Antibody-Antigen) different thresholds are applied.

The anchoring spots mapping procedure searches for sites, in which a protruding residue of one molecule binds in a pocket on the surface of the binding partner. It consists of a geometry based step that detects cavities and sub-cavities on the protein surface, and an energy based step in which amino acid probes are scattered in the vicinity of the cavities. These positions are optimized and their binding energies in the presence of a hypothetical proteins are estimated. We use a new scoring function that takes into consideration solvation effect and dielectric shielding. The algorithm successfully detected known anchoring sites with accurate positioning of the probe and accurate estimate of the binding energy, thereby proving clues for likely protein-protein and protein-peptide interaction sites. Both procedures were tested on a large database of unbound docking models, and were applied in the recent CAPRI predictions.

Katchalski-Katzir, E. (1992). Molecular Surface Recognition: Determination of Geometric Fit Between Proteins and Their Ligands by Correlation Techniques Proceedings of the National Academy of Sciences, 89 (6), 2195-2199 DOI: 10.1073/pnas.89.6.2195
Heifetz, A. (2002). Electrostatics in protein-protein docking Protein Science, 11 (3), 571-587 DOI: 10.1110/ps.26002
Berchanski, A., Shapira, B., & Eisenstein, M. (2004). Hydrophobic complementarity in protein-protein docking Proteins: Structure, Function, and Bioinformatics, 56 (1), 130-142 DOI: 10.1002/prot.20145
Kowalsman, N., & Eisenstein, M. (2009). Combining interface core and whole interface descriptors in postscan processing of protein-protein docking models Proteins: Structure, Function, and Bioinformatics, 77 (2) DOI: 10.1002/prot.22436