基于多层语义与拓扑融合的异质图方法提升药物-靶标相互作用预测性能

doi:10.12122/j.issn.1673-4254.2025.11.12

南方医科大学学报 ›› 2025, Vol. 45 ›› Issue (11): 2394-2404.doi: 10.12122/j.issn.1673-4254.2025.11.12

基于多层语义与拓扑融合的异质图方法提升药物-靶标相互作用预测性能

陈紫豪¹(), 郭延哺¹^,², 宋胜利¹(), 郭全明¹, 周冬明³^,⁴()

^1.郑州轻工业大学软件学院，河南郑州 450001
^2.东南大学江苏省网络群体智能重点实验室，江苏南京 211189
^3.湖南信息学院电子科学与工程学院，湖南长沙 410151
^4.云南大学信息学院，云南昆明 650500

收稿日期:2025-05-21 出版日期:2025-11-20 发布日期:2025-11-28
通讯作者: 宋胜利,周冬明 E-mail:chenzh000523@163.com;2003010@zzuli.edu.cn;zhoudm@ynu.edu.cn
作者简介:陈紫豪，在读硕士研究生，E-mail: chenzh000523@163.com
基金资助:
国家自然科学基金(62403437);国家自然科学基金(62066047);河南省科技攻关项目(242102211039);郑州轻工业大学青年骨干教师培养资助项目(13502010009)

A heterogeneous graph method integrating multi-layer semantics and topological information for improving drug-target interaction prediction

Zihao CHEN¹(), Yanbu GUO¹^,², Shengli SONG¹(), Quanming GUO¹, Dongming ZHOU³^,⁴()

^1.College of Software, Zhengzhou University of Light Industry, Zhengzhou 450001, China
^2.Jiangsu Provincial Key Laboratory of Networked Collective Intelligence, Southeast University, Nanjing 211189, China
^3.School of Electronic Science and Engineering, Hunan University of Information Technology, Changsha 410151, China
^4.School of Information Science and Engineering, Yunnan University, Kunming 650500, China

Received:2025-05-21 Online:2025-11-20 Published:2025-11-28
Contact: Shengli SONG, Dongming ZHOU E-mail:chenzh000523@163.com;2003010@zzuli.edu.cn;zhoudm@ynu.edu.cn
Supported by:
National Natural Science Foundation of China(62403437)

摘要/Abstract

摘要：

目的为解决药物-靶标相互作用预测中存在的高阶语义依赖建模不足、语义路径融合缺乏自适应性及节点特征过平滑等问题，提出一种基于多层语义与拓扑融合的异质图预测方法。方法构建包含药物、蛋白质、副作用、疾病等多类实体的异质图网络，利用图嵌入技术获取低维特征表示。通过自适应元路径搜索模块，自动挖掘语义路径组合，引导高阶语义信息的传播；构建融合多头注意力的语义聚合机制，根据上下文信息自动学习各语义路径的重要性，实现路径间信息的差异化聚合与动态融合；引入结构感知的门控图卷积模块，调控特征传播强度，有效抑制冗余信息，缓解过平滑问题。最终通过内积操作预测药物与靶标之间的相互作用关系。结果本文所提方法在公开数据集上，接收机工作特征曲线下面积（AUC）和精确召回率曲线下面积（AUPRC）分别比现有药物靶标互作用预测方法的平均性能提高了3.4%和2.4%、3.0%和3.8%。结论本文设计的药物-靶标相互作用预测方法可有效提取异质生物网络中复杂的高阶语义和拓扑信息，提升药物-靶标相互作用预测的准确性和稳定性，可为药物靶标的精准发现和复杂疾病的精准治疗提供技术支撑和理论依据。

关键词: 药物-靶标相互作用, 异质网络, 门控机制, 多头注意力机制, 图卷积网络

Abstract:

Objective To develop a heterogeneous graph prediction method based on the fusion of multi-layer semantics and topological information for addressing the challenges in drug-target interaction prediction, including insufficient modeling of high-order semantic dependencies, lack of adaptive fusion of semantic paths, and over-smoothing of node features. Methods A heterogeneous graph network with multiple types of entities such as drugs, proteins, side effects, and diseases was constructed, and graph embedding techniques were used to obtain low-dimensional feature representations. An adaptive metapath search module was introduced to automatically discover semantic path combinations for guiding the propagation of high-order semantic information. A semantic aggregation mechanism integrating multi-head attention was designed to automatically learn the importance of each semantic path based on contextual information and achieve differentiated aggregation and dynamic fusion among paths. A structure-aware gated graph convolutional module was then incorporated to regulate the feature propagation intensity for suppressing redundant information and redcuing over-smoothing. Finally, the potential interactions between drugs and targets were predicted through an inner product operation. Results Compared with existing drug-target interaction prediction methods, the proposed method achieved an average improvement of 3.4% and 2.4%, 3.0% and 3.8% in terms of the area under the receiver operating characteristic curve (AUC) and the area under the precision-recall curve (AUPRC) on public datasets, respectively. Conclusion The drug-target interaction prediction method developed in this study can effectively extract complex high-order semantic and topological information from heterogeneous biological networks, thereby improving the accuracy and stability of drug-target interaction prediction. This method provides technical support and theoretical foundation for precise drug target discovery and targeted treatment of complex diseases.

Key words: drug-target interaction, heterogeneous networks, gated mechanism, multi-head attention mechanism, graph convolutional networks

陈紫豪, 郭延哺, 宋胜利, 郭全明, 周冬明. 基于多层语义与拓扑融合的异质图方法提升药物-靶标相互作用预测性能[J]. 南方医科大学学报, 2025, 45(11): 2394-2404.

Zihao CHEN, Yanbu GUO, Shengli SONG, Quanming GUO, Dongming ZHOU. A heterogeneous graph method integrating multi-layer semantics and topological information for improving drug-target interaction prediction[J]. Journal of Southern Medical University, 2025, 45(11): 2394-2404.

图/表 13

图1 GMADTI总体框架示意图

Fig.1 Overall framework of GMADT.

表1 Luo数据集中节点和边的信息

Tab.1 Information of the nodes and edges in the Luo dataset

Node type	Num	Edge type	Num
Drug	708	Drug-drug (interaction)	10 036
Protein	1512	Drug-drug (interaction)	501 264
Disease	5603	Drug-protein	1923
Side effect	4192	Drug-disease	199 214
		Drug-side effect	80 164
		Protein-disease	1 596 745
		Protein-protein (interaction)	7363
		Protein-protein (similarity)	2 286 144

表2 Zheng数据集中节点和边的信息

Tab.2 Information of the nodes and edges in the Zheng dataset

Node type	Num	Edge type	Num
Drug	1094	Drug-drug	1 196 836
Protein	1556	Drug-drug	11 819
Chemical structure	881	Drug-chemical substructure	133 880
Side effect	4063	Drug-side effect	122 792
Substituent	738	Drug-side effect	20 798
GO term	4098	Protein-GO term	35 980
		Protein-protein	2 421 136

表3 在Luo数据集上与基线方法的对比结果

Tab.3 Comparison results with baseline methods on the Luo dataset

Methods	Luo dataset
Methods	AUC	AUPRC
DTINet	0.879±0.004	0.906±0.003
NeoDTI	0.955±0.003	0.889±0.004
GCN-DTI	0.918±0.005	0.897±0.005
IMCHGAN	0.956±0.004	0.959±0.003
HampDTI	0.928±0.003	0.927±0.005
SGCL-DTI	0.977±0.002	0.976±0.002
GSRF-DTI	0.977±0.002	0.980±0.003
EEG-DTI	0.954±0.003	0.964±0.004
MIDTI	0.978±0.003	0.970±0.002
CE-DTI	0.976±0.002	0.976±0.002
SHGCL-DTI	0.957±0.004	0.958±0.003
AMGDTI	0.977±0.002	0.977±0.002
GMADTI	0.987±0.002	0.987±0.002

表4 在Zheng数据集上与基线方法的对比结果

Tab.4 Comparison results with baseline methods on the Zheng dataset

Methods	Zheng dataset
Methods	AUC	AUPRC
DTINet	0.889±0.004	0.900±0.004
NeoDTI	0.946±0.003	0.846±0.005
GCN-DTI	0.922±0.004	0.914±0.004
IMCHGAN	0.946±0.002	0.929±0.003
EEG-DTI	0.942±0.003	0.941±0.003
SGCL-DTI	0.968±0.002	0.968±0.002
SHGCL-DTI	0.957±0.003	0.961±0.003
CE-DTI	0.972±0.003	0.972±0.002
MIDTI	0.954±0.002	0.949±0.004
AMGDTI	0.973±0.004	0.971±0.002
GMADTI	0.987±0.002	0.983±0.001

图2 在Luo数据集上GMADTI模型及其变体模型比较

Fig.2 Comparison of GMADTI model and its variants on the Luo dataset.

图3 在Zheng数据集上GMADTI模型及其变体模型比较

Fig.3 Comparison of GMADTI model and its variants on the Zheng dataset.

图4 不同注意力头数对GMADTI模型性能的影响

Fig.4 Impact of number of attention heads on the performance of the GMADTI model.

图5 不同学习率对GMADTI模型性能的影响

Fig.5 Impact of learning rates on performance of the GMADTI model.

图6 不同嵌入维度对GMADTI模型性能的影响

Fig.6 Impact of embedding dimensions on performance of the GMADTI model.

图7 药物与靶标节点的 t-SNE 嵌入可视化分布图

Fig.7 t-SNE visualization of drug and target node embeddings.

图8 药物侧不同样本的注意力权重分布热力图

Fig.8 Attention weight heatmap for different samples on the drug side.

图9 靶标侧不同样本的注意力权重分布热力图

Fig.9 Attention weight heatmap for different samples on the target side.

参考文献 34

[1]	Abbasi Mesrabadi H, Faez K, Pirgazi J. Drug-target interaction prediction based on protein features, using wrapper feature selection[J]. Sci Rep, 2023, 13(1): 3594. doi：10.1038/s41598-023-30026-y
[2]	Dehghan A, Abbasi K, Razzaghi P, et al. CCL-DTI: contributing the contrastive loss in drug-target interaction prediction[J]. BMC Bioinformatics, 2024, 25(1): 48. doi：10.1186/s12859-024-05671-3
[3]	刘正美, 魏雪梅, 张俊鹏, 等. 药物分子与靶标蛋白结合亲和力预测研究进展[J].计算机工程与应用, 2024, 60(23): 79-90.
[4]	Khatun MS, Hasan MM, Kurata H. PreAIP: computational prediction of anti-inflammatory peptides by integrating multiple complementary features[J]. Front Genet, 2019, 10: 129. doi：10.3389/fgene.2019.00129
[5]	Ru XQ, Wang LD, Li LH, et al. Exploration of the correlation between GPCRs and drugs based on a learning to rank algorithm[J]. Comput Biol Med, 2020, 119: 103660. doi：10.1016/j.compbiomed.2020.103660
[6]	Ding YJ, Tang JJ, Guo F. Identification of drug-target interactions via multi-view graph regularized link propagation model[J]. Neurocomputing, 2021, 461: 618-31. doi：10.1016/j.neucom.2021.05.100
[7]	Song W, Xu L, Han C, et al. Drug-target interaction predictions with multi-view similarity network fusion strategy and deep interactive attention mechanism[J]. Bioinformatics, 2024, 40(6): btae346. doi：10.1093/bioinformatics/btae346
[8]	Luo Y, Zhao X, Zhou J, et al. A network integration approach for drug-target interaction prediction and computational drug repositioning from heterogeneous information[J]. Nat Commun, 2017, 8(1): 573. doi：10.1038/s41467-017-00680-8
[9]	Zhou D, Xu Z, Li W, et al. MultiDTI: drug-target interaction prediction based on multi-modal representation learning to bridge the gap between new chemical entities and known heterogeneous network[J]. Bioinformatics, 2021, 37(23): 4485-92. doi：10.1093/bioinformatics/btab473
[10]	Qiao G, Wang G, Li Y. Causal enhanced drug-target interaction prediction based on graph generation and multi-source information fusion[J]. Bioinformatics, 2024, 40(10): btae570. doi：10.1093/bioinformatics/btae570
[11]	秦海盈, 赵中英, 李建晖, 等.基于元路径与层次注意力的时序异质信息网络表示学习方法[J].模式识别与人工智能,2021, 34(12): 1093-102.
[12]	Li J, Wang J, Lv H, et al. IMCHGAN: inductive matrix completion with heterogeneous graph attention networks for drug-target interactions prediction[J]. IEEE/ACM Trans Comput Biol Bioinform, 2022, 19(2): 655-65. doi：10.1109/tcbb.2021.3088614
[13]	Su Y, Hu Z, Wang F, et al. AMGDTI: drug-target interaction prediction based on adaptive meta-graph learning in heterogeneous network[J]. Brief Bioinform, 2023, 25(1): bbad474. doi：10.1093/bib/bbad474
[14]	潘柏儒, 丁卫平, 鞠恒荣, 等.基于粗糙图的图卷积神经网络算法[J].模式识别与人工智能, 2023, 35(9): 827-38.
[15]	Wan F, Hong L, Xiao A, et al. NeoDTI: neural integration of neighbor information from a heterogeneous network for discovering new drug-target interactions[J]. Bioinformatics, 2019, 35(1): 104-11. doi：10.1093/bioinformatics/bty543
[16]	Zhao T, Hu Y, Valsdottir LR, et al. Identifying drug-target interactions based on graph convolutional network and deep neural network[J]. Brief Bioinform, 2021, 22(2): 2141-50. doi：10.1093/bib/bbaa044
[17]	Peng J, Wang Y, Guan J, et al. An end-to-end heterogeneous graph representation learning-based framework for drug-target interaction prediction[J]. Brief Bioinform, 2021, 22(5): bbaa430. doi：10.1093/bib/bbaa430
[18]	Zhu Y, Ning C, Zhang N, et al. GSRF-DTI: a framework for drug-target interaction prediction based on a drug-target pair network and representation learning on a large graph[J]. BMC Biol, 2024, 22(1): 156. doi：10.1186/s12915-024-01949-3
[19]	Mukesh M Raghuwanshi Pradnya S Borkar Sachin U Balvir. Node2Vec and machine learning: a powerful Duo for link prediction in social network[J]. Jes, 2024, 20(2s): 639-49. doi：10.52783/jes.1530
[20]	Johnson SJ, Murty MR, Navakanth I. A detailed review on word embedding techniques with emphasis on word2vec[J]. Multimed Tools Appl, 2024, 83(13): 37979-8007. doi：10.1007/s11042-023-17007-z
[21]	Ding Y, Yao Q, Zhao H, et al. Diffmg: Differentiable meta graph search for heterogeneous graph neural networks[C].Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, 2021: 279-88. doi：10.1145/3447548.3467447
[22]	Guo YB, Zhou DM, Ruan XL, et al. Variational gated autoencoder-based feature extraction model for inferring disease-miRNA associations based on multiview features[J]. Neural Netw, 2023, 165: 491-505. doi：10.1016/j.neunet.2023.05.052
[23]	Rasamoelina AD, Adjailia F, Sincak P. A review of activation function for artificial neural network[C]//2020 IEEE 18th World Symposium on Applied Machine Intelligence and Informatics (SAMI). January 23-25, 2020. Herlany, Slovakia. IEEE, 2020: 281-6. doi：10.1109/sami48414.2020.9108717
[24]	Dubey SR, Singh SK, Chaudhuri BB. Activation functions in deep learning: a comprehensive survey and benchmark[J]. Neuroco-mputing, 2022, 503: 92-108. doi：10.1016/j.neucom.2022.06.111
[25]	Deora P, Ghaderi R, Taheri H, et al.On the optimization and generalization of multi-head attention[J]. arXiv preprint arXiv:2310. 12680, 2023.
[26]	郑子瑜, 杨夏颖, 吴圣杰, 等.多特征融合的产时超声胎方位识别模型[J].南方医科大学学报, 2025, 45(7): 1563-70.
[27]	Tian Y, Wang X, Yao X, et al. Predicting molecular properties based on the interpretable graph neural network with multistep focus mechanism[J]. Brief Bioinform, 2023, 24(1): bbac534. doi：10.1093/bib/bbac534
[28]	Wang F, Liu H. Understanding the behaviour of contrastive loss[C].Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021: 2495-504. doi：10.1109/cvpr46437.2021.00252
[29]	Ranjan P, Khan P, Kumar S, et al. $łog$-sigmoid activation-based long short-term memory for time-series data classification[J]. IEEE Trans Artif Intell, 2024, 5(2): 672-83. doi：10.1109/tai.2023.3265641
[30]	Zheng Y, Peng H, Zhang XC, et al. Predicting drug targets from heterogeneous spaces using anchor graph hashing and ensemble learning[C]//2018 International Joint Conference on Neural Networks (IJCNN). July 8-13, 2018. Rio de Janeiro. IEEE, 2018: 1-7. doi：10.1109/ijcnn.2018.8489028
[31]	Srivastava A, Rawat BS, Singh G, et al. A review of optimization algorithms for training neural networks[C]//2023 International Conference on Sustainable Emerging Innovations in Engineering and Technology (ICSEIET). September 14-15, 2023. Ghaziabad, India. IEEE, 2023: 886-90. doi：10.1109/icseiet58677.2023.10303287
[32]	Wang H, Huang F, Xiong Z, et al. A heterogeneous network-based method with attentive meta-path extraction for predicting drug-target interactions[J]. Brief Bioinform, 2022, 23(4): bbac184. doi：10.1093/bib/bbac184
[33]	Li Y, Qiao G, Gao X, et al. Supervised graph co-contrastive learning for drug-target interaction prediction[J]. Bioinformatics, 2022, 38(10): 2847-54. doi：10.1093/bioinformatics/btac164
[34]	Yao KN, Wang XW, Li WN, et al. Semi-supervised heterogeneous graph contrastive learning for drug-target interaction prediction[J]. Comput Biol Med, 2023, 163: 107199. doi：10.1016/j.compbiomed.2023.107199

基于多层语义与拓扑融合的异质图方法提升药物-靶标相互作用预测性能

A heterogeneous graph method integrating multi-layer semantics and topological information for improving drug-target interaction prediction

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 34

相关文章 0

编辑推荐

Metrics

本文评价