利用圖形表示的基因規劃法找尋核醣核酸的共同結構元

Abstract

隨著人們對RNA功能的知識增加，最近對RNA的研究較以往吸引了更多的注意。如同生物體中的其它大分子一樣，RNA的功能取決於它們的結構。由於直接從三級立體結構出發的技術在效用與效率上仍受限制，因此有各式各樣的計算機方法被提出。在這篇論文裡，我們將目標放在RNA的二級結構預測。基於需要被預測的RNA結構的數量不同，計算的方法可被分類為單一序列結構預測以及多重序列結構預測。一般而言，單一序列結構預測被用來尋找一條序列可能的整個二級結構，而多重序列結構預測則是被用來尋找同一RNA家族序列的共同區域性二級結構。目前大部分多重序列結構預測的方法皆侷限於找到相對較短的共同結構元素，它們無法找出較長的共同結構，而這些共同結構很可能在生物學上扮演重要的角色。我們提出了一個多重策略的方法，結合了單一序列結構預測以及多重序列結構預測的方法。藉由使用單一序列結構預測系統的預測結果轉換成我們定義的RNA二級結構的圖形化模型，我們可以提升多重序列結構預測的能力。為了驗證我們系統的效用與效率，我們從Rfam下載數個真實世界裡的RNA家族來做測試，實驗顯示本方法能有不錯的表現。

As the increase of knowledge of RNA functions, the research on RNA has recently attracted more attentions than ever. Like other biopolymers, the functions of RNA are dependent upon their structures. Since the effectiveness and efficiency of ab initio 3D structure determination Technologies are still limited, various computational approaches have been proposed. In this thesis, we are focused on RNA secondary structure prediction. Based on the number of RNA for which to predict the structures, computational methods can be classified as single-sequence prediction and multiple-sequence prediction. In general, single-sequence prediction is aimed to find the probable global secondary structures, and on the other hand, multiple-sequence prediction is aimed to identify the common local secondary structures in a given RNA family. Most of the current approaches to multiple-sequence prediction are limited to finding relatively short common structure elements. As a consequence, they fail to identify those longer common structures that may play important biological roles. We propose a multi-strategy method that combines the advantages of both single-sequence and multiple-sequence prediction. By using the prediction results of single-sequence predictors as the basis to form the graphical models of RNA secondary structures, we can improve the performance in multiple-sequence prediction. To demonstrate the efficiency and effectiveness, we tested our new approach on several real-world RNA families downloaded from Rfam. The experiments showed some promising results.