蛋白质结构类的全序列神经网络预测  

PREDICTION OF PROTEIN STRUCTURAL CLASSES BY USING WHOLE SEQUENCE AND NEURAL NETWORK

作　　者：王焕峰[1] 刘伟[1] 

机构地区：[1]郑州大学生物工程系,郑州河南450001

出　　处：《河南工业大学学报（自然科学版）》2009年第4期65-67,共3页Journal of Henan University of Technology：Natural Science Edition

基　　金：河南省教育厅自然科学研究项目(2007180047)

摘　　要：利用神经网络和蛋白质全序列对蛋白质结构类进行了预测.按照蛋白质序列的长度,分别从508个蛋白质中选择244个蛋白质构建了3个数据集,3个数据集中蛋白质序列的长度变化范围分别为50,100和150个氨基酸.3个数据集的自检结果均为100%.数据集的留一检测结果分别为97.13%,95.90%和95.49%.3个数据集中最好的10倍交叉检验结果为82.38%.结果表明,数据集的序列长度变化范围对预测准确率有很大影响,序列长度变化范围越大准确率越低.The whole protein sequence and a neural network were used to predict protein structural classes. 244 proteins were selected from 508 proteins to respectively construct three datasets according to the length of the protein sequences, wherein the length change ranges of protein sequences in the three datasets were 50, 100 and 150 amino acids, respectively. The self-consistence accuracies of all three datasets are 100% , and the jack - knife test accuracies are 97.13% , 95.90% and 95.49% for the above datasets respectively. The best accuracy of 10-fold cross validation test for three datasets is 82.38%. The results showed that the sequence length range of three datasets plays an important role in the accuracy prediction, and the accuracy is lower when the sequence length range becomes larger.

关 键 词：蛋白质结构类 全序列 神经网络 支持向量机 

分 类 号：TS201[轻工技术与工程—食品科学] O629.73[轻工技术与工程—食品科学与工程]