Abstract: Recognizing the secondary structure of the protein plays an important part in studying on the characteristics and properties of proteins. In the article, the distance matrix is obtained through the three-dimensional coordinates of [Cα]in the protein. For the implicit texture information of the distance matrix, the different directions subband energy and standard deviation are extracted through the level 4 decomposition of the matrix by the double tree complex wavelet transform. Thus, 48 dimensional eigenvectors are got to represent the secondary structural characteristics of proteins. Then, extracted feature is inputted into KNN and SVM classifiers to forecast the secondary structure of protein. Finally, through experimental verification, the dual-tree complex wavelet feature can improve the grain size and direction limitation of traditional feature extraction methods and improve the accuracy of protein secondary structure prediction.

Key words: protein, prediction of secondary structure, dual-tree complex wavelet transform, texture feature, support vector machine

摘要： 识别蛋白质二级结构对于蛋白质的特征和性质研究具有很重要的作用。用[Cα]原子三维空间坐标把蛋白质序列映射为距离矩阵，针对距离矩阵中隐含的纹理信息，用双树复小波变换对矩阵进行4级分解，提取不同方向的子带能量和标准偏差，得到48维特征向量来表示蛋白质的二级结构特征，再将提取的特征输入KNN和SVM分类器分类，通过实验验证，双树复小波特征能改善传统特征提取方法的纹理粒度和方向局限问题，提高蛋白质二级结构的预测准确率。

关键词: 蛋白质, 二级结构预测, 双树复小波变换, 纹理特征, 支持向量机