Fig. 2: Overview of GNN applications for molecules and materials.

a prediction of ADMET properties (adapted with permission from Feiberg et al.¹⁶⁵, Copyright 2020 American Chemical Society), GNNs accounting for environment effects of molecules (reproduced from ref. ¹⁷⁹ with permission from the Royal Society of Chemistry.), GNNs to predict the toxicity of molecules for bees (this illustration was published in ref. ¹⁸⁴, Copyright Elsevier), b RL-based approach for inverse molecular design based on Graph Convolutional Policy Networks (GCPN) (adapted from ref. ¹⁵⁴), c template-free retrosynthesis (adapted from ref. ²²¹), d transferable excited states dynamics (reproduced from ref. ¹⁹⁹ with permission from Springer Nature), coarse graining (reproduced from ref. ¹⁹⁴ with permission from the Royal Society of Chemistry), e explainable GNNs (adapted from ref. ³¹¹), f Crystal GNN to predict methane adsorption volumes in metal organic frameworks (MOFs) (this illustration was published in ref. ²³⁴, Copyright Elsevier), doped structures (this illustration was published in ref. ²⁴⁴, Copyright Elsevier), point defects (adapted with permission from Frey et al.²⁴⁵, Copyright 2020 American Chemical Society) g reactions of Al₂O₃ surface in contact with HF gas (reproduced from ref. ¹⁹⁰ with permission from Springer Nature), GNNs to predict magnetostriction of polycrystalline systems (reproduced from ref. ²⁴⁰ with permission from Springer Nature), h a GNN classifier to predict if a system is in a liquid or a glassy phase only by the positions of the atoms (reproduced from ref. ²⁴⁹ with permission from the Royal Society of Chemistry).