Toward Chemical Accuracy in Predicting Enthalpies of Formationwith General-Purpose Data-Driven Methods

Enthalpies of formation and reaction are important thermodynamicproperties that have a crucial impact on the outcome of chemical transformations. Herewe implement the calculation of enthalpies of formation with a general-purposeANI-1ccx neural network atomistic potential. We demonstrate on a wide range ofbenchmark sets that both ANI-1ccx and our other general-purpose data-driven methodAIQM1 approach the coveted chemical accuracy of 1 kcal/mol with the speed ofsemiempirical quantum mechanical methods (AIQM1) or faster (ANI-1ccx). It isremarkably achieved without specifically training the machine learning parts of ANI-1ccx or AIQM1 on formation enthalpies. Importantly, we show that these data-drivenmethods provide statistical means for uncertainty quantification of their predictions,which we use to detect and eliminate outliers and revise reference experimental data.Uncertainty quantification may also help in the systematic improvement of such data-driven methods.