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

Graph networks are a new machine learning (ML) paradigm that supports both relational reasoning and combinatorial generalization. Here, we develop universal MatErials Graph Network (MEGNet) models for accurate property prediction in both molecules and crystals. We demonstrate that the MEGNet models outperform prior ML models such as the SchNet in 11 out of 13 properties of the QM9 molecule data set. Similarly, we show that MEGNet models trained on ∼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 density functional theory accuracy over a much larger data set. We present…

• NS
  National Science Foundation

  Award: 1053575
• UD
  U.S. Department of Energy

  Awards: ACI-1053575, KC23MP, -AC02-05-CH11231, DE-AC02, AC02-05-CH11231, DE-AC02-05-CH11231, DE-AC02-
• SA
  Samsung Advanced Institute of Technology
• NE
  National Energy Research Scientific Computing Center
• S
  Samsung
• OO
  Office of Science

  Awards: DE-AC02, DE-AC02-05-CH11231 : Materials Project program KC23MP, DE-AC02-05-CH11231
• UO
  University of California, San Diego
• BE
  Basic Energy Sciences

  Awards: DE-AC02, AC02-05-CH11231, KC23MP, DE-AC02-05-CH11231