计算机科学

计算机科学 ›› 2026, Vol. 53 ›› Issue (5): 337-345.doi: 10.11896/jsjkx.250300168

• 人工智能 • 上一篇    下一篇

基于多视图对比与同源特征的药物靶标亲和力预测研究

田鑫1, 朱国胜1, 熊玉然1, 吴悠2   

  1. 1 湖北大学计算机学院 武汉 430062
    2 湖北大学网络空间安全学院 武汉 430062
  • 收稿日期:2025-03-31 修回日期:2025-06-19 发布日期:2026-05-08
  • 通讯作者: 朱国胜(zhuguosheng@hubu.edu.cn)
  • 作者简介:(tianxin109@stu.hubu.edu.cn)
  • 基金资助:
    赛尔网络下一代互联网技术创新项目(NGI I20180411)

Study on Drug Target Affinity Prediction Based on Multi-view Comparison and Homology Information

TIAN Xin1, ZHU Guosheng1, XIONG Yuran1, WU You2   

  1. 1 School of Computer Science, Hubei University, Wuhan 430062, China
    2 School of Cyber Science and Technology, Hubei University, Wuhan 430062, China
  • Received:2025-03-31 Revised:2025-06-19 Online:2026-05-08
  • About author:TIAN Xin,born in 2000,postgraduate.His main research interests include bioinformatics and graph contrast lear-ning.
    ZHU Guosheng,born in 1972,Ph.D,professor,is a member of CCF(No.D4728M).His main research interests include bioinformatics and machine learning.
  • Supported by:
    CERNET Innovation Project(NGI20180411).

PDF (PC)

236

摘要： 现有基于图神经网络的药物-靶标结合亲和力(DTA)预测方法在特征利用与视图融合对齐方面存在不足,主要表现为:1)未能充分建模药物和蛋白质在各自视图中的特征表达方式,导致视图内部特征学习不完整,限制了特征信息的有效利用;2)未能有效对齐不同视图之间的内在关联,限制了跨视图信息的协同作用。为此,提出了基于多视图对比与同源特征的图神经网络模型(MVHGNN)。MVHGNN构建了多视图对比学习框架,分别在药物分子视图与蛋白质视图中采用增强子图拓扑图卷积网络(ESTGCN)和图同构网络(GIN)作为编码器,以学习药物的拓扑结构特征和蛋白质的层次结构特征。同时,利用同源特征整合药物间及蛋白质间的多层次特征,从而增强同一视图内的特征表达能力,提高特征利用率。此外,在药物-靶标亲和力视图中,使用图卷积网络(GCN)提取全局拓扑信息,构建药物-蛋白质的交互表征。进一步地,采用交叉对比学习策略,最大化药物和蛋白质在各自的不同视图下的互信息,提升同类实体的表征一致性,强化跨视图的信息协同。实验结果表明,MVHGNN在两个基准数据集上均表现优越,尤其在Davis数据集上,均方误差(MSE)和修正判定系数(r2m)分别为0.166和0.794,优于现有先进方法。

关键词: 多视图对比学习, 拓扑结构特征, 同源特征, 药物-靶标结合亲和力, 交叉对比学习

Abstract: Existing graph neural network(GNN)-based methods for drug-target affinity(DTA) prediction exhibit notable limitations in feature utilization and view alignment.Specifically,1) they fail to sufficiently model the feature representations of drugs and proteins within their respective views,leading to incomplete intra-view feature learning and restricted exploitation of feature information;2) they are unable to effectively align the intrinsic correlations across different views,thereby limiting cross-view information synergy.To address these challenges,a MVHGNN(Multi-View Hybrid Homogeneous Graph Neural Network ) is proposed.MVHGNN constructs a multi-view contrastive learning framework,employing an ESTGCN(Enhanced Subgraph Topology Graph Convolutional Network) and a GIN(Graph Isomorphism Network) as encoders in the drug molecular view and protein view,respectively,to capture the topological and hierarchical features of drugs and proteins.Furthermore,homology information is integrated to enhance intra-view feature representation and utilization.In the drug-target affinity view,a GCN(Graph Convolutional Network) is used to extract global topological information,enabling the construction of drug-protein interaction representations.A cross-view contrastive learning strategy is further adopted to maximize mutual information between drugs and proteins across different views,enhancing representation consistency and cross-view collaboration.Experimental results demonstrate that MVHGNN achieves superior performance on two benchmark datasets,notably reaching a mean squared error(MSE)of 0.166 and a modified determination coefficient(r2m) of 0.794 on the Davis dataset,outperforming existing state-of-the-art methods.

Key words: Multi-view contrastive learning, Topological features, Homology information, Drug-target affinity(DTA), Cross-view contrastive learning

中图分类号: 

  • TP391
[1]HASSANI K,KHASAHMADI A H.Contrastive multi-viewrepresentation learning on graphs[C]//Proceedings of the 37th International Conference on Machine Learning.PMLR,2020:4116-4126.
[2]CHEN R,CHEN J,GAN X,et al.Multi-View Graph Contrastive Learning for Social Recommendation[J].Scientific Reports,2024,14:22643.
[3]VELIČKOVIĆ P,FEDUS W,HAMILTON W L,et al.Deep graph infomax[J].arXiv:1809.10341,2018.
[4]CHEN T,KORNBLITH S,NOROUZI M,et al.A simpleframework for contrastive learning of visual representations[C]//Proceedings of the 37th International Conference on Machine Learning.PMLR,2020:1597-1607.
[5]GRILL J B,STRUB F,ALTCHÉ F,et al.Bootstrap your own latent:A new approach to self-supervised learning[J].arXiv:2006.07733,2020.
[6]QIU J,CHEN Q,DONG Y.GCC:Graph contrastive coding for graph neural network pre-training[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Disco-very & Data Mining.ACM,2020:1150-1160.
[7]XIONG Z,LIU S,HUANG F.Multi-relational contrastivelearning graph neural network for drug-drug interaction event prediction[C]//Proceedings of the AAAI Conference on Artificial Intelligence.AAAI,2023:5339-5347.
[8]CHENG Z,ZHONG T,ZHANG K.Learning co-ntrastive multi-view graphs for recommendation(student abstract)[C]//Proceedings of the AAAI Conference on Artificial Intelligence.AAAI,2022:12927-12928.
[9]ZHANG L,OUYANG C,LIU Y,et al.Multimodal contrastive representation learning for drug-target binding affinity prediction[J].Methods,2023,220:126-133.
[10]LI Z,REN P,YANG H,et al.TEFDTA:A transformer encoder and fingerprint representation combined prediction method for bonded andnon-bonded drug-target affinities[J].Bioinformatics,2024,40(1):btad778.
[11]GILSON M K,LIU T,BAITALUK M,et al.Binding DB in2015:A public database for medicinal chemistry,computational chemistry and systems pharmacology[J].Nucleic Acids Research,2015,44:D1045-D1053.
[12]KIM S,CHEN J,CHENG T,et al.PubChem 2019 update:Improved access to chemical data[J].Nucleic Acids Research,2018,47:D1102-D1109.
[13]TSUBAKI M,TOMII K,SESE J,et al.Compound-protein interaction prediction with end-to-end learning of neural networks for graphs and sequences[J].Bioinformatics,2018,35(2):309-318.
[14]GHOSH P,HAQUE M A.ResDTA:Predicting drug-targetbinding affinity using residual skip connections[J].arXiv:2303.11434,2023.
[15]VELIČKOVIĆ P,CUCURULL,G,CASANOVA A,et al.Graph Attention Networks[J].arXiv:1710.10903,2018.
[16]WU H J,LIU J K,JIANG T S,et al.AttentionMGT-DTA:A multi-modal drug-target affinity prediction using graph transformer and attention mechanism[J].Neural Networks,2024,169:623-636.
[17]ÖZTÜRK H,ÖZGÜR A,ÖZKI·RIMLI E,et al.DeepDTA:Deep drug target binding affinity prediction[J].Bioinforma-tics,2018,34(17):i821-i829.
[18]NGUYEN T,LE H,QUINN T P,et al.GraphDTA:Predicting drug-target binding affinity with graph neural networks[J].Bioinformatics,2021,37(8):1140-1147.
[19]WANG R Z,ZHANG Y Q,QIN Q Q,et al.Multi-aspect multi-attention fusion of molecular features for drug-target affinity prediction[J].Journal of Computer Applications,2022,42(1):325-332.
[20]ZHAO Q,DUAN G,YANG M,et al.AttentionDTA:Drug-target binding affinity prediction by sequence-based deep learning with attention mechanism[J].IEEE/ACM Transactions on Computational Biology and Bioinformatics,2023,20(2):852-863.
[21]WANG J,XIAO Y,SHANG X,et al.Predicting drug-targetbinding affinity with cross-scale graph contrastive learning[J].Briefings in Bioinformatics,2024,25(1):bbad516.
[22]GUO M H,LIU Z N,MU T J,et al.Beyond Self-Attention:External Attention Using Two Linear Layers for Visual Tasks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2023,45(5):5436-5447.
[23]BOLTON E E,WANG Y,THIESSEN P A,et al.PubChem:Integrated platform of small molecules and biological activities[J].Annual Reports Computational Chemistry,2008,4:217-241.
[24]SMITH T F,WATERMAN M S.Identification of common molecular subsequences[J].Journal of Molecular Biology,1981,147(1):195-197.
[25]OORD A,LI Y,VINYALS O,et al.Representation learningwith contrastive predictive coding[J].arXiv:1807.03748,2018.
[26]DAVIS M I,HUNT J P,HERRGARD S,et al.Comprehensive analysis of kinase inhibitor selectivity[J].Nature Biotechnology,2011,29(11):1046-1051.
[27]TANG J,SZWAJDA A,SHAKYAWAR S,et al.Making sense of large-scale kinase inhibitor bioactivity data sets:A comparative and integrative analysis[J].Journal of Chemical Information and Modeling,2014,54(3):735-743.
[28]HE T,HEIDEMEYER M,GN F,et al.SimBoost:A read-across approach for predicting drug-target binding affinities using gradient boosting machines[J].Journal of Chemistry,2017,9(1):24.
[29]LOVRIĆ M,MOLERO J M,KERN R,et al.PySpark and RDKit:Movingtowards big data in cheminformatics[J].Molecular Informatics,2019,38(6):e1800082.
[30]MICHEL M,HURTADO D M,ELOFSSON A,et al.PconsC4:Fast,accurate and hassle-free contact predictions[J].Bioinformatics,2019,35(15):2677-2679.
[31]JIANG M,LI Z,ZHANG S,et al.Drug-target affinity prediction using graph neural network and contact maps[J].RSC Advances,2020,10(35):20701-20712.
[32]YUAN W,CHEN G,CHEN C,et al.FusionDTA:Attention-based feature polymerizer and knowledge distillation for drug-target binding affinity prediction[J].Briefings inBioinforma-tics,2022,23(1):bbab506.
[33]CHU Z,HUANG F,FU H,et al.Hierarchical graph representation learning for the prediction of drug-target binding affinity[J].Information Sciences,2022,613:507-523.
[34]LI M,GUO Z,WU Y,et al.ViDTA:Enhanced drug-target affi-nity prediction via virtual graph nodes and attention-based feature fusion[J].arXiv:2412.19589,2024.
[35]LI J,YAO L.HCAF-DTA:Drug-target binding affinity prediction with cross-attention fused hypergraph neural networks[J].arXiv:2504.02014,2025.
[36]ROY K,CHAKRABORTY P,MITRA I,et al.Some case stu-dies on application of “r2m” metrics for judging quality of quantitative structure-activity relationship predictions:Emphasis on scaling of response data[J].Journal of Computational Chemistry,2013,34:1071-1082.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市两江新区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发