Pre-training Molecular Graph Representation with 3D Geometry

By Shengchao Liu, Hanchen Wang, Weiyang Liu, Joan Lasenby, Hongyu Guo, Jian Tang. International Conference on Learning Representations, 2021

466 citations recorded by the source corpus.

Molecular graph representation learning is a fundamental problem in modern drug and material discovery. Molecular graphs are typically modeled by their 2D topological structures, but it has been recently discovered that 3D geometric information plays a more vital role in predicting molecular functionalities. However, the lack of 3D information in real-world scenarios has significantly impeded the learning of geometric graph representation. To cope with this challenge, we propose the Graph Multi-View Pre-training (GraphMVP) framework where self-supervised learning (SSL) is performed by leveraging the correspondence and consistency between 2D topological structures and 3D geometric views. GraphMVP effectively learns a 2D molecular graph encoder that is enhanced by richer and more discriminative 3D geometry. We further provide theoretical insights to justify the effectiveness of GraphMVP. Finally, comprehensive experiments show that GraphMVP can consistently outperform existing graph SSL methods.

Subfields: Artificial Intelligence, Computational Theory and Mathematics, Materials Chemistry

View Paper: https://www.semanticscholar.org/paper/b8f816e23ff40d6afabccca2ee4770087ef0ef57

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Anh Tuan Luu
2. Dominique Beaini
3. Ladislav Rampášek
4. Guy Wolf
5. Mikhail Galkin
6. Guolin Ke
7. Vijay Prakash Dwivedi
8. Zhewei Wei

In Computational Theory and Mathematics

1. Tianfan Fu
2. Jianzhu Ma
3. Jian Peng
4. Shengchao Liu
5. Jimeng Sun
6. Jiaqi Guan
7. Zhifeng Gao
8. Hang Zheng

In Materials Chemistry

1. Anh Tuan Luu
2. Dominique Beaini
3. Jianzhu Ma
4. Mikhail Galkin
5. Ladislav Rampášek
6. Guy Wolf
7. Guolin Ke
8. Vijay Prakash Dwivedi

Citations in the corpus

• Pocket2Mol: Efficient Molecular Sampling Based on 3D Protein Pockets (2022, background): “As discussed in the previous section, accurately predicting the 3D structure of small molecules is crucial for studying their physical and chemistry properties (Liu et al., 2021), and more importantly, can determine whether they could actually bind to the 3D pocket.”
• MolGenSurvey: A Systematic Survey in Machine Learning Models for Molecule Design (2022, background): “A lot of follow-up works are built on this, including multi-task learning [60], [68], [69], [70], [71], self-supervised learning [72], [73], [74], [75], [76], and few-shot learning [77], [78], [79].”
• Long Range Graph Benchmark (2022, background): “Molecular datasets are very interesting for the study of Graph Transformers since their properties do not only depend on local graph structure defined by covalent bonds, but inter alia also on long-range interactions that define the 3D folding of the molecules, their surface area, or the energy of t”
• Uni-Mol: A Universal 3D Molecular Representation Learning Framework (2023, methodology): “N-gram [53], PretrainGNN [22], GROVER [11], GraphMVP [26], MolCLR [12], and GEM [13] are pretraining methods.”
• ChEmbed: Enhancing Chemical Literature Search Through Domain-Specific Text Embeddings (2025, unknown): “Other methods represent chemical structures directly, using SMILES strings [21, 22, 23, 24] or molecular graphs [25, 26].”