快速輻射半徑基底函數網路演算法於蛋白質相對溶劑可接觸性預測的應用

Abstract

蛋白質在生物體中一直扮演著很重要的角色，蛋白質被發現的數量及其結構逐年增加。隨著蛋白質的應用越來越廣泛，待解決的課題也就越來越多。例如：蛋白質二級結構預測問題、蛋白質相對溶劑可接觸性預測問題等。目前在蛋白質結構問題的解決上，科學家都是利用X光繞射以及核磁共振 (NMR) 來取得實驗結果。這些方法雖然正確率高，但是相對地所要花費的時間及成本是相當高的。因此利用電腦科學中的機器學習 (Machine learning) 演算法來預測這些問題，相信能夠有效降低實驗與時間成本。

本篇論文，我們利用修改的快速輻射半徑基底函數網路演算法，混合從PSI-BLAST產生的位置加權矩陣，針對蛋白質相對溶劑可接觸性預測問題進行研究。最近歐等人 [10]，發展出快速輻射半徑基底函數網路演算法，是一種較快速且精確設計之網路，應用於蛋白質二級結構預測有顯著的效果。我們的修改的快速輻射半徑基底函數網路演算法，應用於蛋白質相對溶劑可接觸性預測。我們使用五種不同的快速輻射半徑基底函數網路演算法，應用在三態相對溶劑可接觸性預測和二態相對溶劑可接觸性預測。此五種方法包括:(1) 快速輻射半徑基底函數網路演算法、(2) 二階快速輻射半徑基底函數網路演算法、(3) 一般混合快速輻射半徑基底函數網路演算法、(4) 地域趨勢混合快速輻射半徑基底函數網路演算法、以及(5) 全域趨勢混合快速輻射半徑基底函數網路演算法。我們選擇有最佳表現的一般混合快速輻射半徑基底函數網路演算法，做為建議的演算法。我們也將修改的快速輻射半徑基底函數網路演算法的實驗結果，與近幾年的其他方法比較，並且提出我們的方法改進方向的建議。

Proteins have been played an important role in a creature and the numbers of proteins and their structures have been increased with years. Since protein applications are more widely used, there will be a lot of problems to be solved. For example, there are protein secondary structure prediction problem, protein relative solvent accessibility problem and so on. Nowadays, scientists use X-ray diffraction or nuclear magnetic resonance (NMR) to solve the protein structure problem. Although they can achieve high accuracy, it is expensive and long to solve this protein problem. To reduce the time and the costs, it is imperative to use machine learning algorithms to solve this protein problem.

In this thesis, we study protein relative solvent accessibility problem using a modified QuickRBF method combined with a position-specific scoring matrix (PSSM) generated from PSI-BLAST. The QuickRBF method, recently developed by Ou et al. [10], has been applied to protein secondary structure prediction with excellent results. Our modified QuickRBF method is applied on relative solvent accessibility prediction. Five different kinds of QuickRBF approaches are applied on three-state, E, I, and B, and two-state, E, and B, relative solvent accessibility predictions. These five approaches include: (1) QuickRBF, (2) Two-Stage QuickRBF, (3) Common Fusion QuickRBF, (4) Local Tendency Fusion QuickRBF, and (5) Global Tendency Fusion QuickRBF. We recommend the Common Fusion QuickRBF approach which has the best performance as our modified QuickRBF method. We also compare the results of the modified QuickRBF method with other methods in the recent years, and suggest the improvement direction of our approach in the future.