模糊K-最近相鄰點分類法於蛋白質可溶性預測

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。

本篇論文，我們利用修改的模糊K-最近相鄰點法，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Sim等人 [31]，應用模糊K-最近相鄰點法於蛋白質可溶性預測有顯著的效果。我們提出改進之模糊K-最近相鄰點法，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，所得到的實驗結果與近幾年的其它方法比較，有較佳的預測正確率。我們並與歐等人 [52] 所發表的快速輻射半徑基底函數網路演算法做結合。最後，將這兩種方法之結果做資訊融合以有效地提高預測的準確度。六種修正方法包括:(1) 模糊K-最近相鄰點法、(2) 改進的模糊K-最近相鄰點法、(3) 快速輻射半徑基底函數網路演算法、(4) 第一種線性相加合併法、(5) 第二種線性相加合併法、以及(6) 信心指數合併法。在大部分條件表現最佳的情況下，我們建議選擇第二種線性相加合併法。

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.

Using a position-specific scoring matrix (PSSM) generated from PSI-BLAST in this thesis, we develop the modified fuzzy k-nearest neighbor method to predict the protein relative solvent accessibility. By modifying the membership functions of the fuzzy k-nearest neighbor method by Sim et al. [31], has recently been applied to protein solvent accessibility prediction with excellent results. Our modified fuzzy k-nearest neighbor method is applied on three-state, E, I, and B, and two-state, E, and B, relative solvent accessibility predictions, and its prediction accuracy compares favorly with those by the fuzzy k-NN and QuickRBF approaches. At last, we combine the prediction results of modified fuzzy k-nearest neighbor method and QuickRBF approach to improve the performance. Six modification approaches include: (1) Fuzzy K-Nearest Neighbor Method, (2) Modified Fuzzy K-Nearest Neighbor Method, (3) QuickRBF, (4) Linear Combination Fusion 1, (5) Linear Combination Fusion 2, and (6) Reliability Index Fusion. We recommend the Linear Combination Fusion 2 approach which has shown the best performance in most cases.