Документ взят из кэша поисковой машины. Адрес оригинального документа : http://mccmb.belozersky.msu.ru/2015/proceedings/abstracts/126.pdf
Дата изменения: Mon Jun 15 15:36:55 2015
Дата индексирования: Sat Apr 9 23:32:45 2016
Кодировка:

Quantitative comparison of functional properties in protein-protein complexes Anna Hadarovich
Belarusian State University, 220050, Minsk, Belarus, annchameleon@gmail.com

Ivan Anishchenko, Petras J. Kundrotas
Center for Computational Biology, The University of Kansas, Lawrence, Kansas 66047, USA

Alexander V. Tuzikov
United Institute of Informatics Problems, National Academy of S ciences, 220012, Minsk, Belarus

Ilya A. Vakser
Center for Computational Biology and Department of Molecular Biosciences, The University of Kansas, Lawrence, Kansas 66047, USA

Structural characterization of protein-protein interactions (PPI) is essential for understanding life processes at the molecular level. Because of the inherent limitations of the experimental techniques, only a fraction of known PPI has experimentally solved structures. Computational methods are needed to bridge this gap in structural information. Structural modeling of PPI (protein docking) can be roughly divided into: (i) free docking, where sampling of the binding modes is performed regardless of the possible existence of similar experimentally determined structures, and (ii) template-based or comparative docking, where such similar complexes (templates) determine docking predictions. The detection of suitable templates requires search against a diverse library of protein-protein complexes according to some measure reflecting sequence and/or structure similarity between the target and the template. The widely used TM-score [1] has been shown to perform well in the comparative protein docking [2]. However, the performance significantly deteriorates when templates share only moderate structural similarity with the target (TM-score ~ 0.4 0.6) [3]. We propose to complement the TM-score by Gene Ontology (GO) annotations (GO-score), which account for similarity of functional properties of interacting proteins.

The GO-score is based on a hierarchical dictionary (ontology) of the GO-terms (provided by the Gene Ontology Consortium [4]) in three domains: molecular function, biological process, and cellular component. Most studies on GO annotations consider only one domain (molecular function), because of a weaker correlation between the two remaining domains and the sequence or structure similarity [5]. We show that the use of all three ontology domains results in a better selection of protein-protein templates. A number of existing semantic similarity-based algorithms for calculating the GO-score between single proteins [7] were tested on protein-protein complexes. All such scores were based on the concept of the information content, which relateson the significance level of each particular GO-term to the frequency of its occurrence in a reference database (UniProtKB in our case). The algorithm of Schlicker et al. [6] showed superior performance over the other methods, and thus was adopted in this study. We tested the TM-score and the three GO-scores, one for each domain of the ontology, combined in a linear function TM-score + GO-score = +

+ +d * TM-score,

(1)

where ,, , are coefficients reflecting the contributions from the `molecular function,' `biological process,' and `cellular component' domains of the GO and the TM-score, respectively. The coefficient values were obtained on a set of 807 hetero complexes from the library of docking templates [8] by maximizing the number of cases with the rank of a good model (ligand RMSD [LRMSD] from the native structure < 10 е) scored by (1) better than the rank by the TM-score alone.

Figure 1. Ligand RMSD vs. TM-score (left) and the combined TM+GO-score (right).

The scoring function (1) was tested in actual template-based docking [9] on a set of 587 protein-protein complexes from the DOC
KGROUND

resource

(http://dockground.compbio.ku.edu), with better separation of good models compared to the to the TM-score (better clustering of points in the LRMSD < 10 е region). The better performance of function Figure 2. Precision-recall curves for the TM-score and the GO-scores for molecular function, biological process and cellular component. (1) was confirmed by the precision-recall curves shown in Figure 2 (precision as the fraction of good models among all models, and recall as the fraction of good

models recovered at a given threshold among all good models). At the same time, the GOscores based on a single GO domain performed worse than the TM-score alone (smaller areas under the curves in Figure 2). Among the three GO domains the molecular function domain performs best, close to the TM-score. An example of significantly improved ranking for the good model by the scoring function (1) is in Figure 3. For the target complex of ADPribosylation factor 1 and its binding protein GGA1 (1j2j), the best model with L-RMSD 9.4 е was Figure 3. Example of a protein-protein complex with an improved template selection. The protein-protein target is 1j2j subunits A and B. built based on the template complex of AtVSP9a and AtRABF2b proteins (2efd). This model was ranked 91 by the TM-score and 6 by the function (1).

Our results suggest that functional attributes of protein-protein complexes can be formally compared, enhancing the conventional sequence- and structure-based measures by afunctional term. The proposed similarity measure based on GO-terms can be used in comparative modeling of protein-protein complexes by improving the choice of templates for the model building. In the future, we intend to conduct a thorough analysis of the GO-scores and their combination with TM-score. References: 1. Y. Zhang, J. Skolnick (2004) Scoring Function for Automated Assessment of Protein Structure Template Quality, Proteins, 57:702710. 2. Y. Zhang, J. Skolnick (2005) TM-align: a protein structure alignment algorithm based on the TM-score, Nucleic Acids Research, 33:23022309. 3. J. Negroni, R. Mosca, P. Aloy (2014) Assessing the Applicability of Template-Based Protein Docking in the Twilight Zone, Structure, 22: 13561362. 4. The Gene Ontology Consortium (2013) Gene Ontology Annotations and Resources, Nucleic Acids Research, 41:D530D535. 5. P. Lord et al. (2003) Investigating semantic similarity measures across the Gene Ontology: the relationship between sequence and annotation, Bioinformatics, 19:12751283. 6. A. Schlicker et al. (2006) A new measure for functional similarity of gene products based on Gene Ontology, BMC Bioinformatics, 7:302. 7. F. M. Couto, M. J. Silva, P. M. Coutinho (2007) Measuring semantic similarity between Gene Ontology terms, Data & Knowledge Engineering, 61:137-152. 8. I. Anishchenko, P.J. Kundrotas, A.V. Tuzikov, I.A. Vakser (2014) Structural templates for comparative protein docking, Proteins, doi: 10.1002/prot.24736. 9. P.J. Kundrotas, Z.W. Zhu, J. Janin, I.A. Vakser (2012) Templates are available to model nearly all complexes of structurally characterized proteins,Proceedings of the National Academy of Sciences of the United States of America, 109: 9438-9441.