Hybrid localized graph kernel for machine learning energy‐related properties of molecules and solids

Abstract Nowadays, the coupling of electronic structure and machine learning techniques serves as a powerful tool to predict chemical physical properties broad range systems. With aim improving accuracy predictions, large number representations for molecules solids applications has been developed. In this work we propose novel descriptor based on notion molecular graph. While graphs are largely employed in classification problems cheminformatics or bioinformatics, they not often used regression problem, especially energy‐related properties. Our method is local decomposition atomic environments hybridization two kernel functions: graph contribution that describes pattern Coulomb label encodes finer details geometry. The new energy predictions condensed phase systems demonstrated by considering popular QM7 BA10 datasets. These examples show hybrid localized outperforms traditional approaches such as, example, smooth overlap positions matrices.