《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (4): 1344-1353.DOI: 10.11772/j.issn.1001-9081.2025040505

• 前沿与综合应用 • 上一篇    

基于结构网络协同特征与网格注意力增强KAN的药物靶标相互作用预测

豆旭梦1, 解滨1,2,3(), 张朝晖1,2,3, 赵振刚1, 段菡煜1, 郭澳磊1   

  1. 1.河北师范大学 计算机与网络空间安全学院,石家庄 050024
    2.河北省网络与信息安全重点实验室(河北师范大学),石家庄 050024
    3.供应链大数据分析与数据安全河北省工程研究中心(河北师范大学),石家庄 050024
  • 收稿日期:2025-05-08 修回日期:2025-08-08 接受日期:2025-08-11 发布日期:2025-08-15 出版日期:2026-04-10
  • 通讯作者: 解滨
  • 作者简介:豆旭梦(2001—),女,河北邯郸人,硕士研究生,主要研究方向:机器学习、生物信息学
    张朝晖(1969—),女,河北乐亭人,副教授,博士,CCF会员,主要研究方向:机器学习、图像识别
    赵振刚(1986—),男,河北武安人,实验师,博士,主要研究方向:图像处理、生物信息学
    段菡煜(2001—),女,河北石家庄人,硕士研究生,主要研究方向:机器学习、生物信息学
    郭澳磊(1999—),男,河北邯郸人,硕士研究生,主要研究方向:机器学习、遥感图像处理。
  • 基金资助:
    国家自然科学基金资助项目(62476078);中央引导地方科技发展基金资助项目(236Z0104G)

Drug-target interaction prediction based on structure-network collaborative features and grid-attention enhanced Kolmogorov-Arnold network

Xumeng DOU1, Bin XIE1,2,3(), Zhaohui ZHANG1,2,3, Zhengang ZHAO1, Hanyu DUAN1, Aolei GUO1   

  1. 1.College of Computer and Cyber Security,Hebei Normal University,Shijiazhuang Hebei 050024,China
    2.Hebei Provincial Key Laboratory of Network and Information Security (Hebei Normal University),Shijiazhuang Hebei 050024,China
    3.Hebei Provincial Engineering Research Center for Supply Chain Big Data Analytics and Data Security (Hebei Normal University),Shijiazhuang Hebei 050024,China
  • Received:2025-05-08 Revised:2025-08-08 Accepted:2025-08-11 Online:2025-08-15 Published:2026-04-10
  • Contact: Bin XIE
  • About author:DOU Xumeng, born in 2001, M. S. candidate. Her research interests include machine learning, bioinformatics.
    ZHANG Zhaohui, born in 1969, Ph. D., associate professor. Her research interests include machine learning, image recognition.
    ZHAO Zhengang, born in 1986, Ph. D., experimentalist. His research interests include image processing, bioinformatics.
    DUAN Hanyu, born in 2001, M. S. candidate. Her research interests include machine learning, bioinformatics.
    GUO Aolei, born in 1999, M. S. candidate. His research interests include machine learning, remote sensing image processing.
  • Supported by:
    National Natural Science Foundation of China(62476078);Central Government Guided Local Science and Technology Development Fund(236Z0104G)

摘要:

药物?靶标相互作用(DTI)预测是药物发现与再利用的关键任务,它的难点是融合多源异构特征以全面表征药物与靶标间的复杂关联。针对传统方法依赖单一数据源的问题和建模复杂非线性关系方面的不足,提出一种基于结构?网络协同特征与网格注意力增强的KAN(Kolmogorov-Arnold Network)的DTI预测方法(SNKDTI)。首先,设计结构与网络协同的特征提取策略:在药物表示方面,融合分子指纹与图嵌入方法以量化化学结构;在靶标表示方面,结合传统物理化学编码与预训练模型提取序列特征;同时,引入药物?疾病关联和蛋白质相互作用等异构网络,基于重启随机游走(RWR)算法提取网络的拓扑特征,并利用去噪自编码器(DAE)压缩特征,从而融合药物与靶标的结构与网络信息;其次,构建异质生物信息网络(HBIN),使用图卷积网络(GCN)传播特征,并提出一种网格注意力增强的KAN(GA-KAN),以通过引入多组可学习的B样条基函数网格与注意力机制,实现多个非线性映射模块的自适应组合,从而增强模型的表达能力与输入适应性;最后,使用梯度提升决策树(GBDT)构建端到端预测框架。在公开数据集上的对比实验结果表明,SNKDTI的特征曲线下面积(AUC)、精确度?召回率曲线下面积(AUPR)和F1-score相较于对应指标的最优基准方法分别提升了0.81%、1.36%和3.29%。以上验证了SNKDTI在准确性、鲁棒性和泛化能力方面均有显著提升,可为新药靶标筛选提供高效工具。

关键词: 药物?靶标相互作用预测, 结构?网络协同特征, Kolmogorov-Arnold网络, 图神经网络, 注意力机制

Abstract:

Drug-Target Interaction (DTI) prediction is a key task in drug discovery and repurposing. The challenge lies in how to integrate multi-source heterogeneous features to characterize the complex relationships between drugs and targets comprehensively. To address the shortcomings of traditional methods that rely on a single data source and model complex nonlinear relationships in a low-quality way, a DTI prediction method based on Structure-Network collaborative features and grid-attention enhanced KAN (Kolmogorov-Arnold Network) (SNKDTI) was proposed. Firstly, a feature extraction strategy based on structure and network collaboration was designed: for drug representation, molecular fingerprints were fused with graph embedding methods to quantify chemical structures; for target representation, traditional physicochemical encoding was combined with pre-trained models to extract sequence features. Meanwhile, heterogeneous networks such as drug-disease associations and protein-protein interactions were introduced, network topological features were extracted by using the Random Walk with Restart (RWR) algorithm, and the features were compressed by using a Denoising AutoEncoder (DAE), so as to integrate structural and network information of drugs and targets. Secondly, a Heterogeneous Biological Information Network (HBIN) was constructed to carry out feature propagation by using a Graph Convolutional Network (GCN). Additionally, a Grid-Attention enhanced KAN (GA-KAN) was proposed, which introduced multiple learnable B-spline basis function grids and attention mechanisms to achieve adaptive combinations of multiple nonlinear mapping modules, thereby enhancing the model’s expressive power and input adaptability. Finally, a Gradient Boosting Decision Tree (GBDT) was used to build an end-to-end prediction framework. Experimental results of comparing the proposed method and benchmark methods on public datasets show that SNKDTI achieves improvements of 0.81%, 1.36%, and 3.29% in Area Under the receiver operating characteristic Curve (AUC), Area Under the Precision-Recall curve (AUPR), and F1-score, respectively, over the best-performing benchmark methods. The above prove that SNKDTI enhances accuracy, robustness, and generalization ability significantly, providing an efficient tool for new drug target screening.

Key words: Drug-Target Interaction (DTI) prediction, structure-network collaborative feature, KAN (Kolmogorov-Arnold Network), Graph Neural Network (GNN), attention mechanism

中图分类号: