在已知的蛋白质结构研究方法基础上,提出将多分类问题转化成一对多的二分类问题,来预测蛋白质的未知结构。训练多个单分类器进行分类;选用后向传播(Back Propagation,BP)神经网络作为分类预测模型;以伪氨基酸作为网络输入特征;选用Chou提出的蛋白质数据集;实验数据采用全交叉验证(Jackknife)。结果表明:此法能够提高蛋白质三级结构预测的准确率。
Based on the known research method on protein structural class, a method of converting multi-classification problems into one to many two-category problems to predict unknown protein structure is presented. Training multiple single classifiers to classify the protein structural class, choosing back propagation neural network as the model of classifying and predicting the protein structural class, taking the feature of pseudo-amino acid (PseAA) composition as the input of the neural network, validating all experimental data by jackknife, and taking the two hundred and four protein sequences studied by Chou as the dataset, it is shown that our method can improve the predictive accuracy rate.