基于广义回归神经网络的蛋白质二级结构预测

在生化实验中,关于优化蛋白质预测问题,由于采集的信息、参数、选取和设置等优化处理存在随机性,限制了蛋白质二级结构预测精确度。为解决上述问题,针对广义回归神经网络学习速率快、网络稳健的特点,提出基于广义回归神经网络预测蛋白质二级结构的方法。鉴于编码方式对预测精度有重要影响,首先基于5位编码和不同的滑动窗口构建多个广义回归神经网络预测器对蛋白质二级结构进行预测,取得了较好的结果。并采用富含生物进化信息的序列谱(Profile)编码构建输入向量、并针对不同大小的滑动窗口设置多个spread值重新创建广义回归神经网络预测器,大大提高了预测精确度,仿真结果证明了预测模型的有效性和可行性,为预测提供了有效方法。

Protein Secondary Structure Prediction Based on Generalized Regression Neural Network

In order to improve the accuracy of protein secondary structure prediction,for the fast learning rate and the stability of general regression neural network(GRNN),the paper proposed a method for predicting protein secondary structure based on GRNN.As encoding methods play an important part in prediction accuracy,this paper first constructed GRNN predictors for protein secondary structure prediction based on 5-bit encoding and multiple different sliding windows,and achieved preferable results.To further improve the precision,Profile encoding with abundant biological evolution information was used to build input vector.Then,different spread values based on different sliding windows were set up to re-create GRNN predictors,which greatly increased the prediction accuracy.The results show the availability and feasibility of prediction models.