计算机科学
分子图
图形
药物发现
代表(政治)
理论计算机科学
交互网络
分子描述符
交互信息
图论
数据挖掘
可视化
财产(哲学)
化学信息学
计算生物学
人工智能
药品
生物系统
数量结构-活动关系
分子动力学
机器学习
图形数据库
空间分析
外部数据表示
生物网络
分子生物物理学
小分子
分子模型
作者
Zimai Zhang,Yujie Qi,Xi Zhou,Xiaobo Zhu,Lina Hu,Pengwei Hu
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2025-12-15
卷期号:: 7293-7300
标识
DOI:10.1109/bibm66473.2025.11356731
摘要
Understanding molecular structures and their hierarchical relationships with biological entities is crucial for precise drug modeling and discovery. In this study, we proposed the Molecular Hierarchical Graph Network (MHGN), a unified framework that connects the drug molecular graph structure constructed based on atom and bond information with the drug heterogeneous interaction graph. MHGN first encodes detailed chemical features-including atom type, valence, chirality, bond type, aromaticity, and spatial topology-into expressive molecular graph representations. We then constructed a heterogeneous interaction graph for drug discovery to enable bidirectional information exchange between chemical information within drug molecules and non-drug entities, thereby linking atomic-level knowledge with biological information between target or miRNAs. Beyond molecular graph representations, MHGN also leverages the latest molecular visual modality, integrating 3D molecular visual information through a late-fusion strategy to enhance representation completeness. Experimental results on drug-target interaction (DTI) and drug-miRNA interaction (DMI) prediction tasks demonstrate that MHGN outperforms recent state-of-the-art baselines across ROC-AUC, AUPR, and F1-score metrics, highlighting the superiority of MHGN in enabling atom-level information exchange. Furthermore, MHGN generalizes effectively to molecular property regression, achieving an RMSE of 0.6633 and an$R^{2}$of 0.8735 on the ESOL solubility dataset. In summary, MHGN provides a powerful graph-based model for data-driven drug discovery, enabling efficient information exchange between drugs and target or miRNAs at the atomic level.
科研通智能强力驱动
Strongly Powered by AbleSci AI