Identification of Protein鈥揚rotein Interactions by Detecting Correlated Mutation at the Interface

详细信息    查看全文

• 作者：Fei Guo ; Yijie Ding ; Zhao Li ; Jijun Tang
• 刊名：Journal of Chemical Information and Modeling
• 出版年：2015
• 出版时间：September 28, 2015
• 年：2015
• 卷：55
• 期：9
• 页码：2042-2049
• 全文大小：411K
• ISSN：1549-960X

文摘

Protein鈥損rotein interactions play key roles in a multitude of biological processes, such as de novo drug design, immune response, and enzymatic activity. It is of great interest to understand how proteins in a complex interact with each other. Here, we present a novel method for identifying protein鈥損rotein interactions, based on typical co-evolutionary information. Correlated mutation analysis can be used to predict interface residues. In this paper, we propose a non-redundant database to detect correlated mutation at the interface. First, we construct structure alignments for one input protein, based on all aligned proteins in the database. Evolutionary distance matrices, one for each input protein, can be calculated through geometric similarity and evolutionary information. Then, we use evolutionary distance matrices to estimate correlation coefficient between each pair of fragments from two input proteins. Finally, we extract interacting residues with high values of correlation coefficient, which can be grouped as interacting patches. Experiments illustrate that our method achieves better results than some existing co-evolution-based methods. Applied to SK/RR interaction between sensor kinase and response regulator proteins, our method has accuracy and coverage values of 53% and 45%, which improves upon accuracy and coverage values of 50% and 30% for DCA method. We evaluate interface prediction on four protein families, and our method has overall accuracy and coverage values of 34% and 30%, which improves upon overall accuracy and coverage values of 27% and 21% for PIFPAM. Our method has overall accuracy and coverage values of 59% and 63% on Benchmark v4.0, and 50% and 49% on CAPRI targets. Comparing to existing methods, our method improves overall accuracy value by at least 2%.