Graph Neural Network Model Accelerates Biomass Adsorption Energy Prediction on Iron-group Hydrotalcite Electrocatalysts

Iron-group layered double hydroxides (LDH) have demonstrated excellent biomass electrooxidation performance. However, the development of these materials relies on extensive experiments and high computational costs. Therefore, we developed a graph neural network (GNN) (named GALE-Net 2.0) for predicting the adsorption energies in the electrocatalytic reaction of 5-hydroxymethylfurfural (HMF). A data set of the adsorption energies of organic molecules on the LDH was constructed. The GNN model predicted that the 1:2 CoNi-doped LDH catalyst would demonstrate excellent HMF electrooxidation performance. The calculation time was reduced from 24 h with the density functional theory (DFT) calculations to 1 h with the GALE-Net 2.0. The mean absolute error of the GNN model was 0.17 eV, which is consistent with the accuracy of the DFT calculations. Moreover, the model showed some generality as it successfully predicted the adsorption energy of furan derivatives. Our results suggest that GALE-Net 2.0 can accelerate the design of electrocatalysts.