基于CNN与LSTM模型的蛋白质二级结构预测  

Protein secondary structure prediction based on CNN and LSTM models

作　　者：王剑 成金勇[1] 赵志刚 鹿文鹏[1] WANG Jian1 , CHENG Jinyong1 , ZHAO Zhigang2, LU Wenpeng1(1. College of Information, Qilu University of Technology（ Shandong Academy of Sciences ）, Jinan 250353, China, 2.Shandong Computer Science Center（ National Supercomputer Center in Jinan） Qilu University of Technology （ Shandong Academy of Sciences ）, Jinan 250101, China)

机构地区：[1]齐鲁工业大学(山东省科学院)信息学院,济南250353 [2]齐鲁工业大学(山东省科学院)山东省计算中心(国家超级计算济南中心),济南250101

出　　处：《生物信息学》2018年第2期130-136,共7页Chinese Journal of Bioinformatics

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

摘　　要：蛋白质结构的预测在理解蛋白质结构组成和蛋白质的生物学功能有重要意义,而蛋白质二级结构预测是蛋白质结构预测的重要环节。当PSSM位置特异性进化矩阵被广泛应用于将蛋白质初级结构序列编码作为输入样本后,每个残基可以被表示成二维空间的数据平面,由此文中尝试利用卷积神经网络对其进行训练。文中还设计了另一种卷积神经网络,利用长短记忆网络感知了CNN最后卷积特征面的横向特征和纵向特征后连同卷积神经网络的全连接共同完成分类,最后用ensemble方法对两类卷积神经网络模型进行了整合,最终ensemble方法中包含两类卷积神经网络的六个模型,在CB513蛋白质数据集测得的Q3结果为77.2。The prediction of protein structure is of great significance in understanding the structure and the biological function of proteins. The prediction of protein secondary structure is an important part of protein structure prediction. When PSSM position-specific evolution matrix is widely used to encode the primary sequence of a protein,and used as input sample,each residue can be represented as a two-dimensional data plane. Therefore,a convolutional neural network can be adopted as a model to train them. In this paper,we also designed another type of CNN in which LSTM were used to perceive the features of CNN last convolution feature maps both horizontally and vertically,and completed classification collaboratively with the fully-connected neural elements of convolution model. Finally,an ensemble method was adopted to integrate these two types of CNN models. This designed ensemble method includes six models of these two types of CNN. The Q3 accuracy obtained from CB513 is 77.2.

关 键 词：卷积神经网络 长短记忆网络 蛋白质二级结构预测 Ensemble方法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]