Graph Neural Network With Local Frame for Molecular Potential Energy Surface

Xiyuan Wang, Muhan Zhang
Proceedings of the First Learning on Graphs Conference, PMLR 198:19:1-19:30, 2022.

Abstract

Modeling molecular potential energy surface is of pivotal importance in science. Graph Neural Networks have shown great success in this field. However, their message passing schemes need special designs to capture geometric information and fulfill symmetry requirement like rotation equivariance, leading to complicated architectures. To avoid these designs, we introduce a novel local frame method to molecule representation learning and analyze its expressivity. Projected onto a frame, equivariant features like 3D coordinates are converted to invariant features, so that we can capture geometric information with these projections and decouple the symmetry requirement from GNN design. Theoretically, we prove that given non-degenerate frames, even ordinary GNNs can encode molecules injectively and reach maximum expressivity with coordinate projection and frame-frame projection. In experiments, our model uses a simple ordinary GNN architecture yet achieves state-of-the-art accuracy. The simpler architecture also leads to higher scalability. Our model only takes about \textdollar 30\%\textdollar inference time and \textdollar 10\%\textdollar GPU memory compared to the most efficient baselines.

Cite this Paper

BibTeX
@InProceedings{pmlr-v198-wang22d, title = {Graph Neural Network With Local Frame for Molecular Potential Energy Surface}, author = {Wang, Xiyuan and Zhang, Muhan}, booktitle = {Proceedings of the First Learning on Graphs Conference}, pages = {19:1--19:30}, year = {2022}, editor = {Rieck, Bastian and Pascanu, Razvan}, volume = {198}, series = {Proceedings of Machine Learning Research}, month = {09--12 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v198/wang22d/wang22d.pdf}, url = {https://proceedings.mlr.press/v198/wang22d.html}, abstract = {Modeling molecular potential energy surface is of pivotal importance in science. Graph Neural Networks have shown great success in this field. However, their message passing schemes need special designs to capture geometric information and fulfill symmetry requirement like rotation equivariance, leading to complicated architectures. To avoid these designs, we introduce a novel local frame method to molecule representation learning and analyze its expressivity. Projected onto a frame, equivariant features like 3D coordinates are converted to invariant features, so that we can capture geometric information with these projections and decouple the symmetry requirement from GNN design. Theoretically, we prove that given non-degenerate frames, even ordinary GNNs can encode molecules injectively and reach maximum expressivity with coordinate projection and frame-frame projection. In experiments, our model uses a simple ordinary GNN architecture yet achieves state-of-the-art accuracy. The simpler architecture also leads to higher scalability. Our model only takes about \textdollar 30\%\textdollar inference time and \textdollar 10\%\textdollar GPU memory compared to the most efficient baselines.} }
Endnote
%0 Conference Paper %T Graph Neural Network With Local Frame for Molecular Potential Energy Surface %A Xiyuan Wang %A Muhan Zhang %B Proceedings of the First Learning on Graphs Conference %C Proceedings of Machine Learning Research %D 2022 %E Bastian Rieck %E Razvan Pascanu %F pmlr-v198-wang22d %I PMLR %P 19:1--19:30 %U https://proceedings.mlr.press/v198/wang22d.html %V 198 %X Modeling molecular potential energy surface is of pivotal importance in science. Graph Neural Networks have shown great success in this field. However, their message passing schemes need special designs to capture geometric information and fulfill symmetry requirement like rotation equivariance, leading to complicated architectures. To avoid these designs, we introduce a novel local frame method to molecule representation learning and analyze its expressivity. Projected onto a frame, equivariant features like 3D coordinates are converted to invariant features, so that we can capture geometric information with these projections and decouple the symmetry requirement from GNN design. Theoretically, we prove that given non-degenerate frames, even ordinary GNNs can encode molecules injectively and reach maximum expressivity with coordinate projection and frame-frame projection. In experiments, our model uses a simple ordinary GNN architecture yet achieves state-of-the-art accuracy. The simpler architecture also leads to higher scalability. Our model only takes about \textdollar 30\%\textdollar inference time and \textdollar 10\%\textdollar GPU memory compared to the most efficient baselines.
APA
Wang, X. & Zhang, M.. (2022). Graph Neural Network With Local Frame for Molecular Potential Energy Surface. Proceedings of the First Learning on Graphs Conference, in Proceedings of Machine Learning Research 198:19:1-19:30 Available from https://proceedings.mlr.press/v198/wang22d.html.

Related Material