软件导刊2025，Vol.24Issue(9)：26-33,8.DOI:10.11907/rjdk.241998
基于异构符号图神经网络的3D嵌入模型研究
Research on 3D Embedding Model Based on Heterogeneous Symbolic Graph Neural Network
摘要
Abstract
In the field of biomedical science,there are various interaction relationships between drugs and targets,which can be divided us-ing the structural characteristics of graphs and learned and predicted using graph neural network models.The drug target network is composed of numerous nodes,and the characteristics of the nodes are multimodal.At present,most drug target interaction prediction models learn and predict through their own heterogeneous networks combined with single feature information.However,only considering its heterogeneous char-acteristics without learning isomorphic graphs with similar structural features may limit the model's ability to accurately learn and extract relat-ed structures.In addition,a single feature information does not fully reveal the structural details of nodes.To address the above issues,a pre-diction model called 3D-SHGNN is proposed.This model uses supervised classification for link symbol prediction,taking links as samples,fusing the features of drugs and target nodes based on balance theory as the final features of links,using link symbols as sample labels,and in-troducing logistic regression models to train samples to predict the symbol attributes of unknown links.Two different types of symbolic drug tar-get datasets were selected for experiments,and the results showed that the 3D-SHGNN model performed well in AUC,AUPR,F1 value and other indicators on both datasets,especially with AUC reaching 94.3%and 88.6%respectively,verifying its superiority.关键词
药物—靶标相互作用/3D图嵌入/符号图/图神经网络/监督分类/平衡理论Key words
drug-target interaction/3D graph embedding/signed graph/graph neural networks/supervised classification/balance theory分类
信息技术与安全科学引用本文复制引用
肖伟,朱振宇,曾浪,杨为超..基于异构符号图神经网络的3D嵌入模型研究[J].软件导刊,2025,24(9):26-33,8.基金项目
国家自然科学基金面上项目(62072174) （62072174）
