GraphGIM: rethinking molecular graph contrastive learning via geometry image modeling
Abstract Background Learning molecular representations is crucial for accurate drug discovery. Using graphs to represent molecules is a popular solution, and many researchers have used contrastive learning to improve the generalization of molecular graph representations. Results In this work, we rev...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
BMC
2025-07-01
|
| Series: | BMC Biology |
| Subjects: | |
| Online Access: | https://doi.org/10.1186/s12915-025-02249-0 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849238525778591744 |
|---|---|
| author | Chaoyi Li Hongxin Xiang Wenjie Du Tengfei Ma Haowen Chen Xiangxiang Zeng Lei Xu |
| author_facet | Chaoyi Li Hongxin Xiang Wenjie Du Tengfei Ma Haowen Chen Xiangxiang Zeng Lei Xu |
| author_sort | Chaoyi Li |
| collection | DOAJ |
| description | Abstract Background Learning molecular representations is crucial for accurate drug discovery. Using graphs to represent molecules is a popular solution, and many researchers have used contrastive learning to improve the generalization of molecular graph representations. Results In this work, we revisit existing graph-based contrastive methods and find that these methods have limited diversity in the constructed sample pairs, resulting in insufficient performance gains. To alleviate the above challenge, we propose a novel molecular graph contrastive learning method via geometry image modeling, called GraphGIM, which enhances the diversity between sample pairs. GraphGIM is pre-trained on 2 million 2D graphs and multi-view 3D geometry images through contrastive learning. Furthermore, we find that as the convolutional layers process the image becomes deeper, the information of feature maps gradually changes from global molecular-level information (molecular scaffolds) to local atomic-level information (molecular atoms and functional groups), which provides chemical information at different scales. Therefore, we propose two variants of GraphGIM, called GraphGIM-M and GraphGIM-P, which fuse feature maps of different scales in the image using a weighted strategy and a prompt-based strategy, respectively. Conclusions Extensive experiments show that GraphGIM and its two variants outperform state-of-the-art graph contrastive learning methods on eight molecular property prediction benchmarks from MoleculeNet and achieve competitive results with state-of-the-art methods. The code is available at https://github.com/cyli029/GraphGIM . |
| format | Article |
| id | doaj-art-367f4098cd1d41ada62c9117fc3aa671 |
| institution | Kabale University |
| issn | 1741-7007 |
| language | English |
| publishDate | 2025-07-01 |
| publisher | BMC |
| record_format | Article |
| series | BMC Biology |
| spelling | doaj-art-367f4098cd1d41ada62c9117fc3aa6712025-08-20T04:01:35ZengBMCBMC Biology1741-70072025-07-0123111210.1186/s12915-025-02249-0GraphGIM: rethinking molecular graph contrastive learning via geometry image modelingChaoyi Li0Hongxin Xiang1Wenjie Du2Tengfei Ma3Haowen Chen4Xiangxiang Zeng5Lei Xu6College of Computer Science and Electronic Engineering, Hunan UniversityCollege of Computer Science and Electronic Engineering, Hunan UniversitySchool of Software Engineering, University of Science and Technology of ChinaCollege of Computer Science and Electronic Engineering, Hunan UniversityCollege of Computer Science and Electronic Engineering, Hunan UniversityCollege of Computer Science and Electronic Engineering, Hunan UniversitySchool of Electronic and Communication Engineering, Shenzhen Polytechnic UniversityAbstract Background Learning molecular representations is crucial for accurate drug discovery. Using graphs to represent molecules is a popular solution, and many researchers have used contrastive learning to improve the generalization of molecular graph representations. Results In this work, we revisit existing graph-based contrastive methods and find that these methods have limited diversity in the constructed sample pairs, resulting in insufficient performance gains. To alleviate the above challenge, we propose a novel molecular graph contrastive learning method via geometry image modeling, called GraphGIM, which enhances the diversity between sample pairs. GraphGIM is pre-trained on 2 million 2D graphs and multi-view 3D geometry images through contrastive learning. Furthermore, we find that as the convolutional layers process the image becomes deeper, the information of feature maps gradually changes from global molecular-level information (molecular scaffolds) to local atomic-level information (molecular atoms and functional groups), which provides chemical information at different scales. Therefore, we propose two variants of GraphGIM, called GraphGIM-M and GraphGIM-P, which fuse feature maps of different scales in the image using a weighted strategy and a prompt-based strategy, respectively. Conclusions Extensive experiments show that GraphGIM and its two variants outperform state-of-the-art graph contrastive learning methods on eight molecular property prediction benchmarks from MoleculeNet and achieve competitive results with state-of-the-art methods. The code is available at https://github.com/cyli029/GraphGIM .https://doi.org/10.1186/s12915-025-02249-0Contrastive learningMolecular representation learningComputer vision |
| spellingShingle | Chaoyi Li Hongxin Xiang Wenjie Du Tengfei Ma Haowen Chen Xiangxiang Zeng Lei Xu GraphGIM: rethinking molecular graph contrastive learning via geometry image modeling BMC Biology Contrastive learning Molecular representation learning Computer vision |
| title | GraphGIM: rethinking molecular graph contrastive learning via geometry image modeling |
| title_full | GraphGIM: rethinking molecular graph contrastive learning via geometry image modeling |
| title_fullStr | GraphGIM: rethinking molecular graph contrastive learning via geometry image modeling |
| title_full_unstemmed | GraphGIM: rethinking molecular graph contrastive learning via geometry image modeling |
| title_short | GraphGIM: rethinking molecular graph contrastive learning via geometry image modeling |
| title_sort | graphgim rethinking molecular graph contrastive learning via geometry image modeling |
| topic | Contrastive learning Molecular representation learning Computer vision |
| url | https://doi.org/10.1186/s12915-025-02249-0 |
| work_keys_str_mv | AT chaoyili graphgimrethinkingmoleculargraphcontrastivelearningviageometryimagemodeling AT hongxinxiang graphgimrethinkingmoleculargraphcontrastivelearningviageometryimagemodeling AT wenjiedu graphgimrethinkingmoleculargraphcontrastivelearningviageometryimagemodeling AT tengfeima graphgimrethinkingmoleculargraphcontrastivelearningviageometryimagemodeling AT haowenchen graphgimrethinkingmoleculargraphcontrastivelearningviageometryimagemodeling AT xiangxiangzeng graphgimrethinkingmoleculargraphcontrastivelearningviageometryimagemodeling AT leixu graphgimrethinkingmoleculargraphcontrastivelearningviageometryimagemodeling |