利用演化式計算找尋蛋白質結構之相似結構元

Abstract

同一分類蛋白質中的一群相似結構元，不但可以描述此分類的結構特性，對於蛋白質功能的分析也扮演著重要的角色。本論文提出了一個以演化式計算為基礎，並結合分群演算法的模型，來找尋蛋白質結構之相似結構元。透過演化運算子的運作，及適應性函數的導引，研究模型的確能自動尋找到結構非常相似的相似結構元。本研究以著名的蛋白質功能區域 – EF-hand所屬的蛋白質分類為實驗對象，驗證本研究對於找尋同一分類蛋白質的相似結構元能力。本研究比以往的方法更自動、解決或規避了一些以往研究方法所面臨的困難處，並同時兼顧了合理的時間。本研究結果亦提出了一群結構相當相似的相似結構元，部分相似結構元還兼具了蛋白質功能及分類上的意義。

Large-scale functional annotations of proteins can be greatly aided by the identification of a set of motifs that characterize a specific SCOP fold. In this study we describe a new computational method, PRODEC (PROtein structural motifs Detector using Evolutionary Computing), to automatically discover structure motifs in proteins. A key feature of PRODEC is that each PRODEC motif is a duo consisting of a sequence pattern and a structure pattern. PRODEC, based on genetic computation, begins with an initial population of random motifs. Through the evolutionary process, PRODEC iteratively improves the statistical significance of motifs by modifying their configurations. To evaluate each new pattern, a novel scoring function is developed that measure motifs conserved both in 1D sequence and 3D structure. At last, we provide a modified clustering method to refinement the final results. By evolutionary computing operators and clustering, PRODEC can automatically connect short or subtle motifs together and then extend them to longer ones. Tests indicate that PRODEC can successfully detect fold-specific conserved, flexible, and longer structural motifs. Comparing with conventional methods, PRODEC has better performance in finding flexible and long motifs than other motif discovery methods.