蛋白質二級結構預測

Abstract

蛋白質二級結構提供關於蛋白質結構預測重要資訊。有許多預測方法使用這樣的編碼來預測蛋白質二級結構：以待測位置為中心，加上前後胺基酸序列形成一個視窗。在這個視窗裡，以㈰種胺基酸在每一個位置出現的頻率為序列特徵表示碼。我們使用了新的序列特徵表示碼：在視窗裡，以Chou and Fasman 所計算出的胺基酸二及結構趨向係數為序列特徵表示碼。另外，為了考慮序列上相鄰較遠的胺基酸之間的作用力對二及結構的影響，我們也設計了七種化學作用力加入編碼裡。結果顯示，我們的預測正確率達到75%, 特別在β摺板的預測有明顯的提高。並且我們的序列特徵表示法比起其他方法精簡了約七倍，所以在時間上也相對快了兩倍以上。

The prediction of secondary structure usually makes use of a sliding sequence window of a specific length (usually 9-15 amino acid residues) of a protein. In this work, we showed that a novel reduced representation of the input sequence vector can give superior results to the existing method based on the usual binary representation of protein sequences. Our approach is based on the multi-class support vector machines that make use of reduced feature vectors consisting of the homology-weighted information of amino acids. Despite the relatively smaller size of our feature vectors, our approach gives prediction accuracy of 75%, which is better than the 73% of the well-known PHD approach.