哺乳动物蛋白质氧链糖基化位点的PCA神经网络预测  

Prediction of O-Glycosylation Sites in Mammalian Protein by Neural Network Based on Principal Component Analysis (PCA)

作　　者：杨雪梅[1] 李志西[2] 

机构地区：[1]咸阳师范学院数学与信息科学学院,陕西咸阳712000 [2]西北农林科技大学食品科学与工程学院,陕西杨凌712100

出　　处：《西北农业学报》2010年第3期16-20,共5页Acta Agriculturae Boreali-occidentalis Sinica

基　　金：陕西省科技攻关项目(2005K03-G03);陕西省教育厅科学研究计划项目(09JK809)

摘　　要：运用主成分分析与人工神经网络相结合的方法预测哺乳动物蛋白质氧链糖基化位点。以各种窗口长度（N=5,7,9,11,21,31,41,51）的哺乳动物蛋白质序列为研究对象,首先分析蛋白质序列的结构特点,以主成分分析法（PCA）提取主成分,降低样本向量的维数,消除样本向量各分量之间的相关性;然后用一个含单隐层并且输出层只有一个神经元的BP神经网络对氧链糖基化位点进行预测,以确定蛋白质序列中的丝氨酸残基或苏氨酸残基是否糖基化;最后用Matthews相关系数对预测结果进行评价。结果表明：①糖基化蛋白质中P、S、T和A的含量比非糖基化蛋白质中的含量高,S在N端和C端附近都易糖基化,而T在N端附近易糖基化;②提出的预测方法准确快速,预测准确率达65%~92.5%,Matthews相关系数在0.35~0.73之间。The O-Glycosylation sites in mammalian protein sequence were predicted by principal component analysis（PCA） and artificial neural network.The mammalian protein sequences with different window sizes（N=5,7,9,11,21,31,41,51） were researched.Firstly,the correlation of the sample vectors was removed and the dimension of the sample vectors was decreased by PCA,then the O-Glycosylation sites in protein were predicted by a layered neural network.In the back propagation（BP） neural network,one output unit gave the prediction whether a particular site of serine or threonine was glycosylated.The experimental results were evaluated by the Matthews correlation coefficient（C） and showed that： 1.The content of proline,serine,threonine and alanine in O-glycosylated protein was higher than those in nonglycosylated protein.The serine near N or C terminus was easily glycosylated and the threonine near N terminus was easily glycosylated;2.The proposed method was faster and more accurate,the prediction accuracy was about 65%~92.5% and the Matthews correlation coefficient（C） was 0.35~0.73.

关 键 词：哺乳动物蛋白质 主成分分析 预测 BP神经网络 糖基化 Matthews相关系数 

分 类 号：Q51[生物学—生物化学] TP183[自动化与计算机技术—控制理论与控制工程]