Localized Coulomb Descriptors for the Gaussian Approximation Potential
This work provides a more efficient and accurate machine learning potential for computational chemistry, particularly for biomolecular datasets like QM7, QM7b, and GDB9, though it is incremental as it builds upon existing Coulomb matrix and Gaussian approximation potential approaches.
The authors tackled the problem of predicting atomization energies and other atomic properties for molecules using machine learning potentials, achieving chemical accuracy on larger molecules than those in the training set, with improved prediction accuracy and computational cost compared to similar methods.
We introduce a novel class of localized atomic environment representations, based upon the Coulomb matrix. By combining these functions with the Gaussian approximation potential approach, we present LC-GAP, a new system for generating atomic potentials through machine learning (ML). Tests on the QM7, QM7b and GDB9 biomolecular datasets demonstrate that potentials created with LC-GAP can successfully predict atomization energies for molecules larger than those used for training to chemical accuracy, and can (in the case of QM7b) also be used to predict a range of other atomic properties with accuracy in line with the recent literature. As the best-performing representation has only linear dimensionality in the number of atoms in a local atomic environment, this represents an improvement both in prediction accuracy and computational cost when considered against similar Coulomb matrix-based methods.