ANI/EFP: Modeling\nLong-Range Interactions in ANI Neural\nNetwork with Effective Fragment Potentials

Deep learning Neural Networks (NN) have been developed\nin the field\nof molecular modeling for the purpose of circumventing the high computational\ncost of quantum-mechanical calculations while rivaling their accuracies.\nAlthough these networks have found great success, they generally lack\nthe ability to accurately describe long-range interactions, which\nmakes them unusable for extended molecular systems. Herein, we provide\na method for partially retraining the deep learning general-use neural\nnetwork ANI, in which the long-range interactions are represented\nvia atomic electrostatic potentials. The electrostatic potentials,\ngenerated with polarizable effective fragment potentials (EFP), are\nused as an additional input feature for the network. This new ANI/EFP\nnetwork can predict solute–solvent interaction energies on\na trained data set with a kcal/mol accuracy. It also shows promise\nin predicting the interaction energies of a solute in solvent environments\nthat have not been included in a training data set. The proposed protocol\ncan be taken as an example and further developed, leading to highly\naccurate and transferable neural network potentials capable of handling\nlong-range interactions and extended molecular systems.