MoIUNet++：自适应粒度显式子结构与互作感知分子表示学习

打开文本图片集

中图分类号：064 doi： 10.1016/j.actphy.2025.100209

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

Fing Xu1 ， Zhiwei Yang 2*， Sirui wu 3， Wu Su 1， Lizhuo Wang1，Deyu Meng 4，5，* ， Jiangang Long 1，* 1 ， ， 710049， Province， . 2 ， ， 710049， Province， . 3 ，Ltd， 710o77， Province， . 4 Research Institute for Mathematical ， ， 710049， Province， . 5 ， ， ， Province， .

Abstract：Molecular representation learning is a critical task in AI-driven drug development. While graph neural networks （GNNs） have demonstrated strong performance gained widespread adoption in this field， efficiently extracting explicitly analyzing functional groups remains a challenge. To addressthis 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， incorporates a substructure masking explainer for quantitative fragment analysis.We evaluated MolUNet++ on tasks including molecular property prediction， drug-drug interaction （DDI） prediction， drug-target interaction （DTI） prediction. Experimental results demonstrate that MolUNet ++ not only outperforms traditional GNN models in predictive performance but also exhibits explicit， intuitive， chemically logical interpretability.This capability provides valuable insights tools for researchers in drug design optimization.

Key Words：Molecular representation learning;GNN; Structure identification; Adaptive granularity

1引言

分子表示在连接生物信息学、化学信息学、材料科学以及尤其是药物发现等多个科学领域与机器学习之间，起着至关重要的桥梁作用[1-3]。（剩余26709字）