基于 RBF 神经网络的蛋白质二级结构预测

Protein Secondary Structure Prediction Based on Radial Basis Function Neural Network

蛋白质二级结构对于研究其功能具有重要作用。采用主成分分析方法对氨基酸的基本物化属性及其二级结构倾向性进行降维降噪处理,使用径向基神经网络对蛋白质二级结构进行预测。主成分分析使得之前 20 ×12 矩阵变为 20 ×4 矩阵,极大地减少了神经网络输入端的维数。在仿真过程中,当窗口大小为 21,扩展函数为 7 时,预测精确度达到了 71. 81%。实验结果表明 RBF 神经网络可以有效的用于蛋白质二级结构的预测。

Protein secondary structure is important to its function research. In this paper, principal component anal- ysis was used to reduce the dimensions and noises in the basic physical and chemical properties of amino acids and secondary structure orientation. Then radial basis neural networks was used to predict the protein secondary struc- ture. Principal component analysis, which turned the previous 20 × 12 matrix into a 20 × 4 matrix, greatly reduced the input dimensions of the network. During the simulating process, the prediction accuracy was reached 71. 81% when training window was 21 and spread was 7. The results showed that the RBF neural network was an effective method in protein secondary structure prediction.