基于残差网络的蛋白质超二级结构图像分类

Image classification of protein super secondary structure based on residual network

为解决SVM、Bayes、RNN(recurrent neural network)等传统算法在蛋白质结构分类任务中精度低的问题,提出一种基于残差网络的蛋白质超二级结构图像分类方法。将PDB(protein data bank)和SCOP(structural classification of proteins)数据库中的4类蛋白质超二级结构3D模型转化为14角度拍摄的2D图像,针对每类图像,通过残差网络单元进行深度特征提取和优化,利用神经网络模型训练,将验证精度最高的模型保存下来并进行测试。实验结果表明,分类精度达到了90.2%,验证了模型的可行性和算法的有效性。

To solve the problem of low accuracy of traditional algorithms such as SVM,Bayes,RNN(recurrent neural network)in protein structure classification task,a protein super secondary structure image classification method based on residual network was proposed.Four kinds of the 3D proteins model of super secondary structure in PDB(protein data bank)and SCOP(structural classification of proteins)datasets were transformed into 2D images taken from 14 angles.For each kind of images,the depth features were extracted and optimized by residual network unit,and then trained by neural network model.The model with the highest verification accuracy was saved and tested.Experimental results show that the classification accuracy reaches 90.2%,which verifies the feasibility of the model and the effectiveness of the algorithm.