物理化学学报 >> 2026, Vol. 42 >> Issue (5): 100209.doi: 10.1016/j.actphy.2025.100209

论文 上一篇    下一篇

MolUNet++：自适应粒度显式子结构与互作感知分子表示学习

徐凡丁1, 杨志伟2,*(), 武思睿3, 苏武1, 王力卓1, 孟德宇4,5,*(), 龙建刚1,*()   

  1. 1 西安交通大学生命科学与技术学院, 陕西 西安 710049
    2 西安交通大学物理学院, 陕西 西安 710049
    3 中国移动通信集团陕西有限公司, 陕西 西安 710077
    4 西安交通大学数学与统计学院, 陕西 西安 710049
    5 河南大学数学与统计学院, 河南 郑州 475004
  • 收稿日期:2025-08-22 修回日期:2025-10-21 录用日期:2025-10-22 发布日期:2026-01-23
  • 通讯作者: Email: jglong@xjtu.edu.cn (龙建刚)dymeng@mail.xjtu.edu.cn (孟德宇)yzws-123@xjtu.edu.cn (杨志伟)

MolUNet++: adaptive-grained explicit substructure and interaction aware molecular representation learning

Fanding Xu1, Zhiwei Yang2,*(), Sirui Wu3, Wu Su1, Lizhuo Wang1, Deyu Meng4,5,*(), Jiangang Long1,*()   

  1. 1 School of Life Science and Technology, Xi'an Jiaotong University, Xi'an 710049, Shaanxi Province, China
    2 School of Physics, Xi'an Jiaotong University, Xi'an 710049, Shaanxi Province, China
    3 China Mobile Communications Group Shaanxi Co., Ltd, Xi'an 710077, Shaanxi Province, China
    4 Research Institute for Mathematics and Mathematical Technology, Xi'an Jiaotong University, Xi'an 710049, Shaanxi Province, China
    5 School of Mathematics and Statistics, Henan University, Zhengzhou 475004, Henan Province, China
  • Received:2025-08-22 Revised:2025-10-21 Accepted:2025-10-22 Published:2026-01-23
  • Contact: Email: jglong@xjtu.edu.cn (Jiangang Long)dymeng@mail.xjtu.edu.cn (Deyu Meng)yzws-123@xjtu.edu.cn (Zhiwei Yang)

RichHTML

12

PDF

102

赞

补充材料

1

摘要：

分子表示学习是人工智能驱动药物研发中的关键任务。尽管图神经网络(GNN)在该领域已表现出优异性能并被广泛应用，但高效提取并显式解析官能团仍是一项挑战。为此，我们提出了MolUNet++模型，该模型通过分子边收缩池化(Molecular Edge Shrinkage Pooling，MESPool)实现分层子结构提取，利用嵌套式UNet框架进行多粒度特征融合，并结合子结构掩蔽解释器实现分子片段的定量分析。我们在分子性质预测、药物-药物相互作用(Drug-Drug Interaction，DDI)预测及药物-靶标相互作用(Drug-Target Interaction，DTI)预测等任务上对MolUNet++进行了评估。实验结果表明，MolUNet++不仅在预测性能上优于传统GNN模型，同时展现出显式、直观且符合化学逻辑的可解释性，为药物设计与优化领域的研究者提供了有价值的启示与工具。

关键词: 分子表示学习, 图神经网络, 结构识别, 自适应粒度

Abstract:

Molecular representation learning is a critical task in AI-driven drug development. While graph neural networks (GNNs) have demonstrated strong performance and gained widespread adoption in this field, efficiently extracting and explicitly analyzing functional groups remains a challenge. To address this issue, we propose MolUNet++, a novel model that employs Molecular Edge Shrinkage Pooling (MESPool) for hierarchical substructure extraction, utilizes a Nested UNet framework for multi-granularity feature integration, and incorporates a substructure masking explainer for quantitative fragment analysis. We evaluated MolUNet++ on tasks including molecular property prediction, drug-drug interaction (DDI) prediction, and drug-target interaction (DTI) prediction. Experimental results demonstrate that MolUNet++ not only outperforms traditional GNN models in predictive performance but also exhibits explicit, intuitive, and chemically logical interpretability. This capability provides valuable insights and tools for researchers in drug design and optimization.

Key words: Molecular representation learning, GNN, Structure identification, Adaptive granularity