改进的新BRNN网络集成及其二级结构预测的研究  

Study on predicting protein secondary structure based on improved BRNN ensemble

作　　者：林丽玉[1] 黄高昂[1] 左瑞娟[1] 

机构地区：[1]福建师范大学软件学院,福建福州350007

出　　处：《计算机与应用化学》2008年第8期981-984,共4页Computers and Applied Chemistry

基　　金：福建省教育厅资助项目(JB07055).

摘　　要：为提高二级结构预测精度,试用神经网络集成法预测。针对BRNN网络结构复杂、收敛时间长、参数多的缺点,本文提出一种改进的新BRNN网络,删除BRNN左、右子网络的隐层,直接将输入连接到状态层。并采用BP改进算法中的弹性算法训练。以90条蛋白质序列共15 377个氨基酸交叉验证,仿真结果表明新网络可以有效地缩短收敛时间,新BRNN集成预测二级结构效果较好。To improve the performance of secondary structure prediction, Bidirectional Recurrent Neural Network （BRNN） Ensemble was applied. Because BRNN has superflous parameters, complicated structure and slow convergence, this paper gave an improved BRNN structure by deleting it＇s right and left hidden-layer and took the RPROP algorithm to train the network. The data of 90 sequences of 15 377 amino acids were used to test the method. The experiment results showed that the improved network could highly increase the convergent speed and improved BRNN ensemble achieved better performance on secondary structure prediction.

关 键 词：BRNN 神经网络集成 二级结构 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]