Prediction of Protein-Protein Interaction Sites from Three-Dimensional Structure

Abstract

理解蛋白質交互作用的機制，對於分析蛋白質功能或是了解蛋白質交互作用網路有著舉足輕重的角色。因此，若能預測蛋白質交互作用的區域(protein-protein interaction sites)，對於解讀蛋白質功能與其反應將很有幫助。在這篇論文中，我們使用機器學習(machine-learning approach)的方法，利用胺基酸組成、蛋白質二級結構、胺基酸相對暴露率(relative solvent accessible surface area)、演化資訊與蛋白質結構等資訊，從蛋白質的三級結構來預測此蛋白質與其他蛋白質發生交互作用的區域。我們在一個蛋白質基準集合上，測試了我們的方法，並且順利地預測了蛋白質交互作用區域。這結果顯示出我們的方法在預測蛋白質交互作用區域上是有用的。

The knowledge of protein-protein interactions is essential for the understanding of protein functions and protein-protein interaction networks. Hence, the capability to identify protein-protein interaction sites is crucial to decipher the reaction mechanisms of protein function. In this work we develop am approach based on machine-learning method to predict protein-protein interaction sites from three-dimensional structure. We have tried a multiple of feature vectors such as amino acid composition, secondary structure, relative solvent accessible surface area, position substitution specific matrix and structural neighboring residues. Our results compare favorably with those of others for a benchmark dataset.