基于优化的卷积神经网络的蛋白质二级结构预测  

PREDICTION OF PROTEIN SECONDARY STRUCTURE BASED ON OPTIMAL CONVOLUTIONAL NEURAL NETWORK

作　　者：赵亚武 刘毅慧[1] Zhao Yawu;Liu Yihui(School of Computer Science and Technology,Qilu University of Technology/Shandong Academy of Sciences,Jinan 250353,Shandong,China)

机构地区：[1]齐鲁工业大学(山东省科学院)计算机科学与技术学院,山东济南250353

出　　处：《计算机应用与软件》2021年第7期147-152,166,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(61375013);山东省自然科学基金项目(ZR2013FM020)。

摘　　要：蛋白质二级结构预测对确定蛋白质的空间结构和功能有极其重要的意义。融合卷积神经网络和贝叶斯优化模型,优化卷积神经网络的网络框架和超参数。采用优化后的卷积神经网络在测试集CASP9、CASP10、CASP11、CASP12、CB513和25PDB上获得了81.99%、81.36%、80.83%、78.10%、84.29%和84.80%的Q3准确率。实验结果表明,相比于其他方法,该模型提高了蛋白质3类二级结构预测的精度。Protein secondary structure prediction is extremely important for determining the spatial structure and function of proteins.The convolutional neural network and Bayesian optimization model were combined to optimize the network framework and hyperparameters of convolutional neural network.The optimized convolutional neural network obtained Q3 accuracy of 81.99%,81.36%,80.83%,78.10%,84.29%and 84.80%in the test sets CASP9,CASP10,CASP11,CASP12,CB513 and 25 PDB.The experimental results show that compared with other methods,the model improves the accuracy of protein class 3 secondary structure prediction.

关 键 词：蛋白质二级结构预测 贝叶斯优化 卷积神经网络 生物信息学 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]