AnkPlex: algorithmic structure for refinement of near-native ankyrin-protein docking

BMC Bioinformatics

Research Article

Kuntida Kitidee¹  Watshara Shoombuatong²  Chatchai Tayapiwatana³  Tanchanok Wisitponchai³  Vannajan Sanghiran Lee⁴ 

[1]Center of Biomolecular Therapy and Diagnostic, Faculty of Associated Medical Sciences, Chiang Mai University, 50200, Chiang Mai, Thailand

[2]Center for Research and Innovation, Faculty of Medical Technology, Mahidol University, 10700, Bangkok, Thailand

[3]Center of Data Mining and Biomedical Informatics, Faculty of Medical Technology, Mahidol University, 10700, Bangkok, Thailand

[4]Division of Clinical Immunology, Department of Medical Technology, Faculty of Associated Medical Sciences, Chiang Mai University, 50200, Chiang Mai, Thailand

[5]Center of Biomolecular Therapy and Diagnostic, Faculty of Associated Medical Sciences, Chiang Mai University, 50200, Chiang Mai, Thailand

[6]Thailand Center of Excellence in Physics, Commission on Higher Education, 10400, Bangkok, Thailand

[7]Department of Chemistry, Faculty of Science, University of Malaya, 50603, Kuala Lumpur, Malaysia

关键词: Ankyrin-protein complexes;    Near-native docking pose;    Machine learning methods;    Decision tree;    Logistic regression model;    AnkPlex;   

DOI  :  10.1186/s12859-017-1628-6

 received in 2016-06-30, accepted in 2017-04-07,  发布年份 2017

来源: Springer

【 摘 要 】

BackgroundComputational analysis of protein-protein interaction provided the crucial information to increase the binding affinity without a change in basic conformation. Several docking programs were used to predict the near-native poses of the protein-protein complex in 10 top-rankings. The universal criteria for discriminating the near-native pose are not available since there are several classes of recognition protein. Currently, the explicit criteria for identifying the near-native pose of ankyrin-protein complexes (APKs) have not been reported yet.ResultsIn this study, we established an ensemble computational model for discriminating the near-native docking pose of APKs named “AnkPlex”. A dataset of APKs was generated from seven X-ray APKs, which consisted of 3 internal domains, using the reliable docking tool ZDOCK. The dataset was composed of 669 and 44,334 near-native and non-near-native poses, respectively, and it was used to generate eleven informative features. Subsequently, a re-scoring rank was generated by AnkPlex using a combination of a decision tree algorithm and logistic regression. AnkPlex achieved superior efficiency with ≥1 near-native complexes in the 10 top-rankings for nine X-ray complexes compared to ZDOCK, which only obtained six X-ray complexes. In addition, feature analysis demonstrated that the van der Waals feature was the dominant near-native pose out of the potential ankyrin-protein docking poses.ConclusionThe AnkPlex model achieved a success at predicting near-native docking poses and led to the discovery of informative characteristics that could further improve our understanding of the ankyrin-protein complex. Our computational study could be useful for predicting the near-native poses of binding proteins and desired targets, especially for ankyrin-protein complexes. The AnkPlex web server is freely accessible at http://ankplex.ams.cmu.ac.th.

【 授权许可】

CC BY   
© The Author(s). 2017