Equivariant Neural Networks Utilizing Molecular Clusters for Accurate Molecular Crystal Binding Energy Predictions

Ankur Gupta,Wibe de Jong,Miko Stulajter,Yusuf Shaidu,Jeffrey Neaton

DOI: https://doi.org/10.26434/chemrxiv-2023-vk9gh

2023-12-01

Abstract:Equivariant neural networks have emerged as prominent models in advancing the construction of interatomic potentials due to their remarkable data efficiency and generalization capabilities for out-of-distribution data. Here, we expand the utility of these networks to the prediction of crystal structures consisting of organic molecules. Traditional methods for computing crystal structure properties, such as plane-wave quantum chemical methods based on density functional theory (DFT), are prohibitively resource-intensive, often necessitating compromises in accuracy and the choice of exchange-correlation functional. We present an approach that leverages the efficiency, and transferability of equivariant neural networks, specifically Allegro, to predict molecular crystal structure energies at a reduced computational cost. Our neural network is trained on molecular clusters using a highly accurate Gaussian-type orbital (GTO)-based method as the target level of theory, eliminating the need for costly periodic DFT calculations, while providing access to all families of exchange-corelation functionals and post-Hartree Fock methods. The trained model exhibits remarkable accuracy in predicting binding energies, aligning closely with those computed by plane-wave based DFT methods, thus representing significant cost reductions. Furthermore, the Allegro network was seamlessly integrated with the USPEX framework, accelerating the discovery of low-energy crystal structures during crystal structure prediction.

Chemistry

What problem does this paper attempt to address?

The problem discussed in this paper is how to predict the binding energy of molecular crystals more accurately and efficiently. Traditional plane wave methods based on density functional theory (DFT) are widely used to calculate the structural properties of crystals, but they have high computational requirements, limiting their compromise on accuracy and choice of exchange-correlation functionals. The paper proposes a method using equivariant neural networks, particularly Allegro, which predicts the energy of molecular crystal structures by training on molecular cluster data, reducing the computational cost and allowing access to all types of exchange-correlation functionals and post-Hartree-Fock methods. This method is integrated with the USPEX framework and accelerates the discovery of low-energy structures in crystal structure prediction. The research shows that the proposed model is highly consistent with the results of plane wave DFT methods in predicting binding energy, significantly reducing computational costs.

Equivariant Neural Networks Utilizing Molecular Clusters for Accurate Molecular Crystal Binding Energy Predictions

Equivariant Neural Networks Utilizing Molecular Clusters for Accurate Molecular Crystal Lattice Energy Predictions

Learning local equivariant representations for large-scale atomistic dynamics

Reliable and Practical Computational Prediction of Molecular Crystal Polymorphs

Accelerated Organic Crystal Structure Prediction with Genetic Algorithms and Machine Learning

Data-efficient machine learning for molecular crystal structure prediction

Crystal structure prediction by combining graph network and optimization algorithm

Organic Crystal Structure Prediction Via Coupled Generative Adversarial Networks and Graph Convolutional Networks

Accelerated prediction of atomically precise cluster structures using on-the-fly machine learning

Predicting Organic Crystal Lattice Energies with Chemical Accuracy

How Accurate Can Crystal Structure Predictions Be for High-Energy Molecular Crystals?

Enhancing crystal structure prediction by combining computational and experimental data via graph networks

Machine-Learned Fragment-Based Energies for Crystal Structure Prediction

Learning Equivariant Non-Local Electron Density Functionals

A Robust Crystal Structure Prediction Method to Support Small Molecule Drug Development with Large Scale Validation and Prospective Studies

Equivariant Graph Attention Networks for Molecular Property Prediction

Lightweight equivariant interaction graph neural network for accurate and efficient interatomic potential and force predictions

A Hybrid Machine Learning Approach for Structure Stability Prediction in Molecular Co-crystal Screenings

Robust and efficient reranking in crystal structure prediction: a data driven method for real-life molecules

Geometric Deep Learning for Molecular Crystal Structure Prediction

Accurate modeling of the potential energy surface of atmospheric molecular clusters boosted by neural networks