E( n ) Equivariant Graph Neural Network for Learning Interactional Properties of Molecules

Kieran Nehil-Puleo,Co D. Quach,Nicholas C. Craven,Clare MCabe,Peter T. Cummings
DOI: https://doi.org/10.1021/acs.jpcb.3c07304
IF: 3.466
2024-01-18
The Journal of Physical Chemistry B
Abstract:We have developed a multi-input E(n) equivariant graph convolution-based model designed for the prediction of chemical properties that result from the interaction of heterogeneous molecular structures. By incorporating spatial features and constraining the functions learned from these features to be equivariant to E(n) symmetries, the interactional-equivariant graph neural network (IEGNN) can efficiently learn from the 3D structure of multiple molecules. To verify the IEGNN's capability to learn...
chemistry, physical
What problem does this paper attempt to address?
This paper aims to address the problem of predicting chemical properties arising from the interaction of different molecular structures. Traditional Quantitative Structure-Property Relationship (QSPR) models, such as linear regression and random forest, have been widely used for their simplicity and interpretability but may not efficiently handle the vast molecular design space. With the development of deep learning, especially Graph Neural Networks (GNNs), they are capable of learning directly from molecular structures, thereby improving prediction efficiency. The paper presents a new multi-input E(n) equivariant graph convolutional model called Interactive Equivariant Graph Neural Network (IEGNN), which combines spatial features and constrains the learned representation to maintain E(n) symmetry, effectively learning from the 3D structures of multiple molecules. To verify the ability of IEGNN in learning interactive properties, the authors tested the model on three molecular datasets, including two newly created datasets, and made them publicly available for future research. IEGNN outperforms previous methods in terms of predicting frictional properties (such as the coefficient of friction) with the lowest average absolute percentage error on three out of four datasets. Additionally, the model demonstrates the ability to predict unknown interaction relationships, such as the frictional properties between differently composed monolayers. The paper also describes how to construct data structures to batch load multi-graph data points and compares them with traditional GNN and random forest models. The study indicates that IEGNN can more accurately predict properties such as frictional force and adhesion force by considering the three-dimensional structure and interactions of the molecules, particularly outperforming previous random forest-based models in predicting the friction coefficient F0. This highlights the importance of considering spatial information between molecules for developing QSPR models.