基于图卷积和双线性注意力网络的药物靶标亲和力预测  

Drug-Target Affinity Prediction Based on Graph Convolution Network and Bilinear Attention Network

作　　者：程竹平 李建华[1] CHENG Zhuping;LI Jianhua(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：《华东理工大学学报（自然科学版）》2024年第4期594-601,共8页Journal of East China University of Science and Technology

基　　金：国家重大新药创制项目(2018ZX09735002)。

摘　　要：药物靶标亲和力预测在药物研发中扮演着重要的角色。针对现有预测方法大多忽略药物分子的二维结构信息、缺乏深层表征融合学习的问题,提出了基于图卷积和双线性注意力网络的药物靶标亲和力预测模型(GBN_DTA)。该模型首先基于多层图卷积神经网络编码药物分子图,同时结合1D-CNN和双向长短期记忆网络(BiLSTM)编码靶标序列;然后使用双线性注意力网络融合编码后的药物和靶标特征,最终获得亲和力预测分数。实验结果表明,该模型在DAVIS和KIBA数据集上的性能均优于其他6种主流方法,有效提升了预测准确率。Drug target affinity prediction plays an important role in drug development.However,most existing prediction methods neglect the two-dimensional structure information of drug molecules and lack deep representation fusion learning.To address this issue,a drug target affinity prediction model(GBN_DTA)based on graph convolutional network and bilinear attention network is proposed.This model first encodes the drug molecule graph using a multi-layer graph convolutional neural network,and simultaneously encodes the target sequence using 1D-CNN and BiLSTM.Then,a bilinear attention network is used to fuse the encoded drug and target features to obtain the affinity prediction score.Experimental results show that the performance of this model on the DAVIS and KIBA datasets is superior to six other mainstream methods,effectively improving the prediction accuracy.

关 键 词：药物靶标亲和力预测 药物研发 图卷积神经网络 双线性注意力网络 深层表征融合 

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]