Oian基于神经网络的蛋白质序列分类研究

提出了基于Levenberg-Marquardt(LM)算法的BP神经网络对蛋白质序列进行家族分类的新方法．该方法采用二肽含量对蛋白质序列进行特征提取，根据影响因子评价特征的相对重要性，用改进的BP神经网络LM优化算法构造一个三层人工神经网络，通过对PIR数据库中三类家族的学习，该网络对未知蛋白质序列分类的准确率分别达到了98．9％．98．1％，97．8％。

Protein Sequences Classification Based on Neural Networks

A new protein sequences classification method is proposed based on BP networks of Levenberg-Marquardt(LM)optimization algorithm. This approach,adopted the bi-peptide content encoding method to extract the occurrences of patterns of two consecutive amino acids(residues)in a protein sequence,using impact factor to evaluate the relative importance of the patterns,and constructed a three-layer feedforward artificial neural networks using the back propagation algorithm which modified by LM algorithm. After training by three protein superfamilies from PIR protein database,the precision of the classification of the unknown protein sequences can respectively achieve 98.9%,98.1%,97.8%.