使用伪氨基酸组成和BP神经网络预测类弹性蛋白多肽的相变温度

Using Pseudo-Amino Acid Composition and BP Neural Network to Predict the Transition Temperature of Elastin-Like Peptides

根据获得的16条ELP序列及相变温度的数据,利用伪氨基酸组成方法提取其序列特征值.将伪氨基酸组成中的相关系数部分作为类弹性蛋白的特征向量,从类弹性蛋白序列出发,利用最小中位方差回归,找出与其序列相关系数的最佳阶数.运用均匀设计法,分别对支持向量机与BP神经网络参数进行优化.结果表明:BP神经网络获得的预测模型最佳,相变温度绝对误差为0.39℃,均方根误差为0.89℃.

Elastin-like peptides（ELP） is one of the multi-peptides which has been widely used.Transition temperature is the most convenient parameters for quantificational description of the ELP properties.It is of great importance to explore the relationship between the transition temperature and the sequence characteristics,the number of Xaa of each monomer and the concentration of ELP.In this article,the best order of the correlation coefficient for pseudo-amino acid composition was obtained by using Least Median of Squares Regression from sequence.The uniform design was used to optimize the running parameters and leave-one out cross-validation was carried out to evaluate the model of back propagation neural network（BPNN） and support vector machines,respectively.The results showed that the predicted model obtained by BPNN was the best,of which the mean absolute error and root mean squared error was 0.39 ℃ and 0.89 ℃,respectively.