利用結構分析預測蛋白質結合區域之研究

Abstract

蛋白質與受質(ligand)之間的結合區域是蛋白質功能的主要作用區域。由於蛋白質的結合區域具有高度的守恆性，使得結合區域可以作為有效辨識蛋白質功能的特殊樣式(motif)，所以預測蛋白質結合區域是藥物設計上的一項重要議題。在電腦輔助製藥上，常利用已知的protein-ligand complex分析其立體結構來預測蛋白質與受質之間結合區域；在蛋白質立體結構分析的過程中，如何決定結構的相似度對結果有決定性的影響，所以一直是一個重要的課題。在本論文中，我們利用結合受質周圍的胺基酸作為樣板，針對結合區域的守恆性而進行立體結構分析，以此設計了一套以區域結構分析蛋白質與受質結合區域的程序，並著重在相似度分析的部分。其中比較了現有的相似度評分方式，並提出一個辨識蛋白質種類的新評分方法。我們提出的程序使用在酵素功能預測上，可使預測的準確度從大約40%提升到55%，同時大幅提高預測結果的confidence。這套工具的結果可以提供給生化學家一些有用的線索進行更進一步的研究。

The protein-ligand binding site is the major active area of protein function. The property of the volume near protein-ligand binding site is highly correlated to protein function, so the prediction of protein-ligand binding site is a fundamental issue of drug design. Local structure analysis is widely employed in computer aid drug design. How to decide the structural similarity is significance during 3D structure analysis processing. In this thesis, a framework of local structure comparison for prediction of protein-ligand binding site has been proposed. We also provided a comparative analysis of several different scoring functions and designed a novel similarity function on discrimination between homologous protein families. The proposed framework is applied to a functional prediction of enzymes and is able to improve the accuracy from about 40% to 55%. The confidence of the prediction also highly rises. The experimental results provide the biochemists with some valuable clues for in-depth studies.