TorchANI: A Free and Open Source PyTorch Based Deep Learning Implementation of the ANI Neural Network Potentials

Abstract

<div>This paper presents TorchANI, a PyTorch based software for training/inference</div><div>of ANI (ANAKIN-ME) deep learning models to obtain potential energy surfaces and</div><div>other physical properties of molecular systems. ANI is an accurate neural network</div><div>potential originally implemented using C++/CUDA in a program called NeuroChem.</div><div>Compared with NeuroChem, TorchANI has a design emphasis on being light weight,</div><div>user friendly, cross platform, and easy to read and modify for fast prototyping, while</div><div>allowing acceptable sacrifice on running performance. Because the computation of</div><div>atomic environmental vectors (AEVs) and atomic neural networks are all implemented</div><div>using PyTorch operators, TorchANI is able to use PyTorch’s autograd engine to automatically compute analytical forces and Hessian matrices, as well as do force training</div><div>without additional codes required.</div>