High-Throughput Screening of Promising Redox-Active Molecules with MolGAT

Redox flow batteries (RFBs) have emerged as a promisingoptionfor large-scale energy storage, owing to their high energy density,low cost, and environmental benefits. However, the identificationof organic compounds with high redox activity, aqueous solubility,stability, and fast redox kinetics is a crucial and challenging stepin developing an RFB technology. Density functional theory-based computationalmaterials prediction and screening is a time-consuming and computationallyexpensive technique, yet it has a high success rate. To speed up thediscovery of new materials with desired properties, machine-learning-basedmodels can be trained on large data sets. Graph neural networks (GNNs)are particularly well-suited for non-Euclidean data and can modelcomplex relationships, making them ideal for accelerating the discoveryof novel materials. In this study, a GNN-based model called MolGATwas developed to predict the redox potential of organic moleculesusing molecular structures, atomic properties, and bond attributes.The model was trained on a data set of over 15,000 compounds withredox potentials ranging from -4.11 to 2.56. MolGAT outperformedother GNN variants, such as the Graph Attention Network, Graph ConvolutionNetwork, and AttentiveFP models. The trained model was used to screena vast chemical data set comprising 581,014 molecules, namely OMDB,QM9, ZINC, CHEMBL, and DELANEY, and identified 23,467 potential redox-activecompounds for use in redox flow batteries. Of those, 20,716 moleculeswere identified as potential catholytes with predicted redox potentialsup to 2.87 V, while 2,751 molecules were deemed potential anolyteswith predicted redox potentials as low as -2.88 V. This workdemonstrates the capabilities of graph neural networks in condensedmatter physics and materials science to screen promising redox-activespecies for further electronic structure calculations and experimentaltesting.