Graph attention neural networks for mapping materials and molecules beyond short-range interatomic correlations

Yuanbin Liu,Xin Liu,Bing-Yang Cao

DOI: https://doi.org/10.1088/1361-648x/ad2584

2024-02-04

Journal of Physics Condensed Matter

Abstract:Bringing advances in machine learning to chemical science is leading to a revolutionary change in the way of accelerating materials discovery and atomic-scale simulations. Currently, most successful machine learning schemes can be largely traced to the use of localized atomic environments in the structural representation of materials and molecules. However, this may undermine the reliability of machine learning models for mapping complex systems and describing long-range physical effects because of the lack of non-local correlations between atoms. To overcome such limitations, here we report a graph attention neural network as a unified framework to map materials and molecules into a generalizable and interpretable representation that combines local and non-local information of atomic environments from multiple scales. As an exemplary study, our model is applied to predict electronic structure properties of metal-organic frameworks (MOFs) which have notable diversity in compositions and structures. The results show that our model achieves the state-of-the-art performance. The clustering analysis further demonstrates that our model leads to a high-level identification of MOFs with spatial and chemical resolution, which would facilitate the rational design of promising reticular materials. Furthermore, the application of our model in predicting the heat capacity of complex nanoporous materials, a critical property in a carbon capture process, showcases its versatility and accuracy in handling diverse physical properties beyond electronic structures.

physics, condensed matter

What problem does this paper attempt to address?

This paper mainly discusses how to use Graph Attention Neural Networks (GANN) to address the limitations in material and molecular modeling, particularly for complex systems and physical effects beyond short-range atomic correlations. Current machine learning methods mostly rely on structural representations of local atomic environments, which may weaken the predictive reliability for systems with long-range interactions. To solve this problem, the paper proposes a unified framework, namely GANN, which can incorporate local and non-local atomic environment information across multiple scales to generate generalizable and interpretable material and molecular representations. GANN is exemplified by its application in predicting electronic structure properties of Metal-Organic Frameworks (MOFs), demonstrating its superior performance. By predicting the band gaps of MOFs, GANN not only achieves state-of-the-art prediction accuracy but also showcases high-resolution spatial and chemical recognition capabilities through cluster analysis, contributing to the rational design of promising ordered materials. Additionally, the paper also showcases the versatility, data efficiency, and accuracy of GANN in predicting the heat capacity of complex nanoporous materials (including MOFs, conjugated organic frameworks, and zeolites), which is a more challenging task. In summary, the paper aims to address how to utilize deep learning, particularly Graph Attention Neural Networks, to create a universal representation method that captures both local and non-local information of materials and molecules, thereby improving the predictive ability for materials with complex structures and long-range interactions.

Graph attention neural networks for mapping materials and molecules beyond short-range interatomic correlations

Directed graph attention neural network utilizing 3D coordinates for molecular property prediction

Large-scale machine-learning-assisted exploration of the whole materials space

Atomistic graph networks for experimental materials property prediction

Predicting Material Properties Using a 3D Graph Neural Network with Invariant Local Descriptors

Graph neural network framework for energy mapping of hybrid monte-carlo molecular dynamics simulations of Medium Entropy Alloys

Combining crystal graphs and domain knowledge in machine learning to predict metal-organic frameworks performance in methane adsorption

Formula graph self-attention network for representation-domain independent materials discovery

Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals

CEGANN: Crystal Edge Graph Attention Neural Network for multiscale classification of materials environment

Molecular substructure graph attention network for molecular property identification in drug discovery

Coarse-grained crystal graph neural networks for reticular materials design

Graph Neural Network Predictions of Metal Organic Framework CO2 Adsorption Properties

Local environment-based machine learning for molecular adsorption energy prediction

Self-supervised Representations and Node Embedding Graph Neural Networks for Accurate and Multi-scale Analysis of Materials

Machine learning with persistent homology and chemical word embeddings improves prediction accuracy and interpretability in metal-organic frameworks

Graph neural networks for materials science and chemistry

Molecular Graph Transformer: Stepping Beyond ALIGNN Into Long-Range Interactions

Quantum Mechanics and Machine Learning Synergies: Graph Attention Neural Networks to Predict Chemical Reactivity

Equivariant Graph Attention Networks for Molecular Property Prediction