蛋白質主鏈結構預測

Abstract

建立蛋白質在空間上的結構有助於瞭解蛋白質的功能與機制。現今可藉由X-ray與NMR的方法來建構蛋白質的三級結構。但這些方法仍需耗費金錢、人力與時間。目前在PDB資料庫中已有三萬多個蛋白質三級結構。若可以利用此資料庫作為知識背景，利用機器學習的方法來預測未知結構的蛋白質序列的結構與功能。這將使得我們可以更容易瞭解蛋白質的功能與機制。 本研究是尋找適合描述蛋白質區域結構的定義，探討由序列預測主鏈結構的可能性。我們採用了兩種不同的定義Ramachandran Plot 及 Protein Blocks來定義一個蛋白質殘基的區域結構。並以支持向量機來預測，得到相當不錯的結果。 最後我們將研究蛋白質區域結構的保守性與區域結構熵的關係。發現在經由不同的蛋白質區域結構定義所的到的結構熵有相當的一致性。

The knowledge of protein structure conformation is useful in understanding the functions and mechanism of proteins. Nowadays, we can use the X-ray or NMR technique to construct the 3-dimensional structure of proteins. But these methods cost lots of time and money. Fortunately, we now have an easier alternative to the knowledge of protein structures. That is, from the increasing amount of data in the protein structure database, such as PDB database. Up to the present, there are 30 thousand records stored in the PDB database. If we use the PDB database as our knowledge and use the machine learning technique to predict the 3-dimensional structure from the sequence of an unknown protein structure. That can let us easier understand the protein’s function and mechanism. This research is to find a suitable definition to describe protein backbone conformation, and to predict backbone conformation from sequence information. We adapt two different definitions of Ramachandran Plot and Protein Blocks. Our results by Support Vector Machines (SVM) get a relatively good performance. Finally, we will study on the relationship between the conservation of protein backbone conformation and structural entropy. We've discovered that there is a consistency in structural entropy by different definitions of protein backbone conformation.