压缩激励卷积神经网络的蛋白质亚细胞定位  

Squeeze and Excitation Convolutional Neural Network for Protein Subcellular Localizatoin

作　　者：唐浩漾 张小媛 钱萌 王燕 TANG Hao-yang;ZHANG Xiao-yuan;QIAN Meng;WANG Yan(College of Automation,Xi'an University of Post and Telecommunications,Xi'an Shanxi 710121,China)

机构地区：[1]西安邮电大学自动化学院,陕西西安710121

出　　处：《计算机仿真》2022年第7期461-465,共5页Computer Simulation

基　　金：国家自然科学基金(21977082);陕西省国际合作交流项目(2017KW-013);西安市科技计划资助项目(201805040YD18CG24);陕西省教育厅专项科技计划(18JK0702)。

摘　　要：针对蛋白质类间差异小导致蛋白质亚细胞定位准确率低、速度慢的问题,提出一种基于压缩激励卷积神经网络的蛋白质亚细胞定位算法。通过搭建结合压缩激励模块的深度卷积神经网络,增强网络提取特征的表达能力,有效提取亚细胞图像中具有区分度的特征,并采用极限学习机分类器进行训练和分类,实现亚细胞的快速准确定位。实验结果表明,上述算法在提升定位速度的基础上能提高亚细胞定位准确率,对UCSF yeast GFP数据集内10种蛋白质图像的亚细胞定位平均准确率达到93.36%,与Deepyeast网络模型相比,其定位平均准确率提高了3.91%,对于相似度较高的三种亚细胞,其定位准确率可分别提高7.61%、10.94%、8.3%。Aiming at the problem of low protein subcellular localization accuracy and slow speed due to small differences between protein classes,a protein subcellular localization algorithm based on compressed excitation convolutional neural network is proposed.By building a deep convolutional neural network combined with a compressed excitation module,the network’s ability to extract features was enhanced to effectively extract features with discrimination in subcellular images,and the extreme learning machine classifier was used for training and classification to achieve rapid and accurate subcellular Positioning.Experimental results show that the algorithm can improve the accuracy of subcellular localization on the basis of improving the positioning speed.The average accuracy of subcellular localization of 10 protein images in the UCSF yeast GFP data set is 93.36%,which is compared with the Deepyeast network model.The average accuracy of its positioning is increased by 3.91%.For the three protein subcellulars with higher similarity,the accuracy of positioning can be increased by 7.61%,10.94%,and 8.3%,respectively.

关 键 词：压缩激励 卷积神经网络 极限学习机 亚细胞定位 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]