基于距离编码-图神经网络的药物靶标作用关系预测  

Predicting Drug Target Interactions Based on Distance Encoding-graph Neural Network

作　　者：翁兴娜 高创 李建华[1] WENG Xing-na;GAO Chuang;LI Jian-hua(School of Information Science and Engineering,East China University of Science and Technology,Shanghai 200237,China)

机构地区：[1]华东理工大学信息科学与工程学院,上海200237

出　　处：《小型微型计算机系统》2023年第12期2663-2670,共8页Journal of Chinese Computer Systems

基　　金：国家重点研发计划项目(2016YFA0502304)资助。

摘　　要：图神经网络方法在药物靶标相互作用的预测任务上效果较好,但在图数据上仍存在无法分辨相同结构拓扑图和图网络中节点特征表达能力受限的问题.本文提出一种基于距离编码-图神经网络(Distance Encoding-Graph Neural Network)的药物靶标作用关系预测方法DEDTI.DEDTI利用图网络中的结构信息对每个药物和靶标节点进行距离编码,使得具有相同拓扑结构的节点可以投影到不同区域,最终在识别网络拓扑结构方面的能力超过一阶WL测试.另外,距离编码使得节点在送入图神经网络训练之前就包含节点属性,而不只是单纯的one-hot编码,提升了图神经网络的性能.在实验数据集中,DEDTI方法的AUC和AUPR均优于其它基准方法.实验结果表明本方法增强了图神经网络在预测药物靶标相互作用方面的能力,并在常用的多个药物数据库上验证了DEDTI预测新药物靶标相互作用的效果.Graph neural network-based drug-target interaction prediction methods have achieved good performance,but still meet some problems such as inability to distinguish the same structural topology graph and limited ability to express node features in the graph network.In this work,a novel drug-target interaction prediction method based on distance encoding-graph neural network is proposed.In this method,the structural information in the graph network is used to encode each drug and target node,allowing nodes with the same topology to be projected to different regions,ultimately surpassing the first-order WL test in identifying the network topology.In addition,node attributes are included in the nodes by distance encoding before they are fed into the graph neural network for training,which improves the performance of the graph neural network.Our method outperforms other five benchmark methods in both the area under curve and the area under precision-recall curve while predicting drug-target interaction on the DrugBank_FDA and Yammanishi_08 datasets.The experimental results indicate that our method have enhanced the predictive ability of graph neural network in predicting drug-target interactions,and validate the effectiveness of our method in predicting new drug-target interaction on multiple widely used drug datasets.

关 键 词：图神经网络 药物靶标相互作用 距离编码 节点表示 网络结构 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]