基于图卷积神经网络的药物-靶标亲和力预测算法  

Drug-Target Binding Affinity Prediction Algorithm Based on Graph Neural Network

作　　者：邓海生[1] 吴逸嘉 杨江海 DENG Haisheng;WU Yijia;YANG Jianghai(Xijing University,Xi'an Shaanxi 710123,China;Nanning Normal University,Nanning Guangxi 530100,China)

机构地区：[1]西京学院,陕西西安710123 [2]南宁师范大学,广西南宁530100

出　　处：《信息与电脑》2022年第21期88-90,94,共4页Information & Computer

基　　金：陕西省教改项目“新文科视域下复合型新闻传播人才数据素养培养模式研究与实践”(项目编号:21BY179);中国高教学会项目“大数据背景下大学生数据素养评价体系研究”(项目编号:22SZJY0415);西京学院校级精品课程《新媒体数据分析及可视化》(项目编号:XJJPKC22010)。

摘　　要：近年来,由于药物开发比较耗时,且常常伴有安全性问题,使用深度学习技术预测药物-靶标亲和力(Drug-Target Affiinity,DTA)成为一个有效的替代方案。基于此,提出采用基于图卷积神经网络的算法来有效捕获药物的结构信息,学习被表示成图的药物特征,使用一维卷积神经网络学习蛋白质特征。将所提出的方法分别在两个基准药物靶标结合亲和数据集上进行了验证,并与已有的先进模型进行比较。In recent years,because drug development is time-consuming and often accompanied by safety problems,using deep learning technology to predict drug target binding affinity has become an effective alternative.Based on this,the algorithm based on graph convolution neural network is proposed to effectively capture the structural information of drugs,learn the drug features represented as a graph,and use one-dimensional convolution neural network to learn the protein features.The proposed method was validated on two benchmark drug target binding affinity datasets and compared with the existing advanced models.

关 键 词：图卷积神经网络(GCN) 注意力机制 药物-靶标亲和力(DTA) 深度学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] R91[自动化与计算机技术—控制科学与工程]