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

Abstract

Graph networks are a new machine learning (ML) paradigm that supports both relational reasoning and combinatorial generalization. Here, we develop, for the first time, universal MatErials Graph Network (MEGNet) models for accurate property prediction in both molecules and crystals. We demonstrate that our MEGNet models significantly outperform prior ML models in 11 out of 13 properties of the QM9 molecule data set. Furthermore, a single-task unified MEGNet model can accurately predict the internal energy at 0 K and room temperature, enthalpy and Gibbs free energy, with temperature, pressure and entropy being global state inputs. Similarly, we show that MEGNet models trained on $\sim 60,000$ crystals in the Materials Project substantially outperform prior ML models in the prediction of the formation energies, band gaps and elastic moduli of crystals, achieving better than DFT accuracy over a much larger data set. Such MEGNet models are highly interpretable, and well-established periodic chemical trends can be extracted from the elemental embeddings. Finally, we demonstrate the transfer learning of elemental embeddings from a property model trained on a larger data set (formation energies) to accelerate the training of property models with smaller amounts of data (band gaps and elastic moduli)