Boosting heterogeneous catalyst discovery by structurally constrained deep learning models

Abstract

The discovery of new catalysts is one of the significant topics of computational chemistry as it has the potential to accelerate the adoption of renewable energy sources. Recently developed deep learning approaches such as graph neural networks open new opportunity to significantly extend scope for modeling novel high-performance catalysts. Nevertheless, the graph representation of a particular crystal structure is not a straightforward task due to the ambiguous connectivity schemes and numerous embeddings of nodes and edges. Here, we present embedding improvement for graph neural networks that has been modified by Voronoi tessellation and is able to predict the energy of catalytic systems within the Open Catalyst Project dataset. The enrichment of the graph was calculated via Voronoi tessellation, and the corresponding contact solid angles and types (direct/indirect) were considered as edges’ features, and Voronoi volumes were used as node characteristics. The auxiliary approach was enriching node representation by intrinsic atomic properties (electronegativity, period, and group position). The proposed modifications allowed us to improve the mean absolute error of the original model, and the final error equals to 651 meV on the Open Catalyst Project dataset and 6 meV/atom on the intermetallics dataset. Also, by the consideration of an additional dataset, we show that a sensible choice of data can decrease the error to values below a physically-based 20 meV/atom threshold.