使用ab initio結構預測方法建構蛋白質功能區結構模型之研究

Abstract

蛋白質是各式各樣生物體的基本單位。為了更了解蛋白質的功能，科學家設法從不同角度來瞭解蛋白質，包括其序列、功能胺基酸、立體結構等等。在生物資訊領域，蛋白質結構的預測一直是一個各界關注的問題。蛋白質結構問題也就是了解蛋白質如何摺疊的問題，透過X-ray、NMR等實驗方法來找結構，雖然準確度較高，但是所耗費之成本也偏高，於是人們想出了用電腦模擬來預測蛋白質結構，其成效雖無法像實驗方式如此精確，但有些結果已是可以被接受的，如二級結構預測的準確度達約80%，足以提供生物學家許多有用的資訊。想要直接從序列就知道蛋白質結構並不是容易的事，曾經有許多結構預測的方式陸續被開發，如同源序列模擬法、摺疊辨識與從頭開始法(ab initio)等等。本論文旨於研究如何使用ab initio結構預測方法建構蛋白質功能區塊之結構模型。我們使用三方面的預測資訊：保留性胺基酸、結構穩定區域(ordered regions)與區塊邊界(domain boundary)來幫助挑選可靠的結構預測結果。在生物演化過程中，功能區塊往往被小心地保留下來。這些序列上的保留性特徵將作為選擇建構功能區塊結構模型的初始序列範圍之重要依據。本論文提出以非穩定性區域之預測預測結果，判斷蛋白質之功能區塊是否有機會自行摺疊成單獨區塊；接著參考區塊邊界預測軟體所提供的序列區段，進行結構預測；最後再以結構預測軟體所輸出之結構間的一致性，判斷所預測之區間範圍是否正確，並輸出最適當之結構模型。我們所設計的方法，預先篩選摺疊區域，事後並依據Rosetta的能量得分作為挑選蛋白質結構的條件，藉此提高蛋白質結構預測準確度，以期能為蛋白質結構預測之研究有所貢獻。

The basic units of all kinds of organisms are proteins. In order to learn more about the protein function, scientists try to understand proteins from different points of view, including protein sequence, protein structure, location of functional residues, and so forth. In the fields of computational biology, the prediction of protein structure has always been the core issue that researchers indefatigably focus on. It has never been easy to tell the protein structure merely through protein sequence; hence methods for predicting the structures are being developed one after another, including approaches based on homology modeling, fold recognition, and ab initio. The issue of determining protein structure is equivalent to the identification of protein folding. Protein structure can be derived from experimental methods such as X-ray crystallography and NMR spectroscopy. Although the accuracy of the experimental methods is higher, their cost is relatively higher and more time consuming as well; researchers therefore come up with the idea of employing theoretical simulation to predict protein structure. The results of computational approaches may not be as precise as that of experimental methods, but some of them are fairly acceptable. For example, the accuracy of secondary structure prediction is about 80%, which provides valuable clues for biologists. This dissertation focuses mainly on how to determine the structure model of a protein functional domain. In the process of biological evolution, important regions are usually conserved. Our method is based on the information derived from three predictors, including the prediction of conserved residues, ordered regions, and domain boundaries. These characteristics of protein sequences are used as the basis for selecting initial functional regions for structure prediction. Furthermore, a structure comparison program is incorporated to evaluate the quality of the estimated region boundary, in order to increase the accuracy of the structure prediction. Results conducted in this thesis show that the proposed method is effectively in identifying the functional regions and delivering satisfied structure model for the proteins of interest.