Hybrid Model of Neural Network and Hidden Markov Model for Protein Secondary Structure Prediction

摘要：

Protein secondary structure prediction is an important step towards understanding how proteins fold in three dimensions. Using multiple sequence alignments, two layers NN-based method gets higher prediction accuracy. But window-based approach in NN-based method has the disadvantage of only considering the local information. Recent analysis by information theory indicates that the correlation between neighboring secondary structures are much stronger than that of neighboring amino acids. So we use a 7-state HMM to replace the second layer network. 496 proteins selected from the dataset CB513 are used in a 7-fold cross validation. The hybrid model appears to be very efficient, with Q3 score of 75.96％ and SOV of 71.27％, more than 0.96％ and 0.45％ above two layers NN-based method. This hybrid model not only captures the local information, but considers the long-distance information. So it can get higher prediction accuracy.

关键词： protein secondary structure prediction hidden markov model artificial neural network

作者: Ou-Yan Shi Hui-Yun Yang Jing Yang Xin Tian

作者单位: Department of Biomedical Engineering, Tianjin Medical University, Tianjin 300070,China;College of Ba Department of Biomedical Engineering, Tianjin Medical University, Tianjin 300070,China College of Basic Medicine, Tianjin Medical University, Tianjin 300070, China

会议类型: 国际会议

会议名称: The 5th International Forum on Post-genome Technologies(5IFPT)(第五届国际后基因组生命科学技术学术论坛)

会议地点: 苏州

会议语种:英文

在线出版日期: 2007-09-10（万方平台首次上网日期，不代表论文的发表时间）