This paper presents TorchANI, a PyTorch based software for training/inference of ANI (ANAKIN-ME) deep learning models to obtain potential energy surfaces and other physical properties of molecular systems. ANI is an accurate neural network potential originally implemented using C++/CUDA in a program called NeuroChem. Compared with NeuroChem, TorchANI has a design emphasis on being light weight, user friendly, cross platform, and easy to read and modify for fast prototyping, while allowing acceptable sacrifice on running performance. Because the computation of atomic environmental vectors (AEVs) and atomic neural networks are all implemented 1 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 without additional codes required. TorchANI is open-source and freely available on

<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>

Community efforts in the computational molecular sciences (CMS) are evolving toward modular, open, and interoperable interfaces that work with existing community codes to provide more functionality and composability than could be achieved with a single program. The Quantum Chemistry Common Driver and Databases (QCDB) project provides such capability through an application programming interface (API) that facilitates interoperability across multiple quantum chemistry software packages. In tandem with the Molecular Sciences Software Institute and their Quantum Chemistry Archive ecosystem, the unique functionalities of several CMS programs are integrated, including CFOUR, GAMESS, NWChem, OpenMM, Psi4, Qcore, TeraChem, and Turbomole, to provide common computational functions, i.e., energy, gradient, and Hessian computations as well as molecular properties such as atomic charges and vibrational frequency analysis. Both standard users and power users benefit from adopting these APIs as they lower the language barrier of input styles and enable a standard layout of variables and data. These designs allow end-to-end interoperable programming of complex computations and provide best practices options by default.

Transferable high dimensional neural network potentials (HDNNPs) have shown great promise as an avenue to increase the accuracy and domain of applicability of existing atomistic force fields for organic systems relevant to life science. We have previously reported such a potential (Schrodinger-ANI) that has broad coverage of druglike molecules. We extend that work here to cover ionic and zwitterionic druglike molecules expected to be relevant to drug discovery research activities. We report a novel HDNNP architecture, which we call QRNN, that predicts atomic charges and uses these charges as descriptors in an energy model that delivers conformational energies within chemical accuracy when measured against the reference theory it is trained to. Further, we find that delta learning based on a semiempirical level of theory approximately halves the errors. We test the models on torsion energy profiles, relative conformational energies, geometric parameters, and relative tautomer errors.

Liquid electrolytes are one of the most important components of Li-ion batteries, which are a critical technology of the modern world. However, we still lack the computational tools required to accurately calculate key properties of these materials (viscosity and ionic diffusivity) from first principles necessary to support improved designs. In this work, we report a machine learning-based force field for liquid electrolyte simulations, which bridges the gap between the accuracy of range-separated hybrid density functional theory and the efficiency of classical force fields. Predictions of material properties made with this force field are quantitatively accurate compared to experimental data. Our model uses the QRNN deep neural network architecture, which includes both long-range interactions and global charge equilibration. The training data set is composed solely of non-periodic density functional theory (DFT), allowing the practical use of an accurate theory (here, ωB97X-D3BJ/def2-TZVPD), which would be prohibitively expensive for generating large data sets with periodic DFT. In this report, we focus on seven common carbonates and LiPF6, but this methodology has very few assumptions and can be readily applied to any liquid electrolyte system. This provides a promising path forward for large-scale atomistic modeling of many important battery chemistries.

Increasing demand for hybrid materials that merge the synthetic and biological areas in drug industries requires in-depth knowledge of the individual components and their contributions to these complexes. Coarse-grained (CG) models developed for proteins and polymers exist, yet there is a lack of understanding of the cross interactions when these two groups of materials integrate to build a complex. In this work, we characterized the nonbonded interactions between poly(ethylene glycol) (PEG) and amino acids in a Martini CG model utilizing state-of-the-art quantum mechanics calculations of interaction energies. The parameter set proposed, was validated by assessing the polymer density in the vicinity of individual amino acids obtained from available all-atomistic molecular dynamic simulations of plasma proteins. Our results revealed the necessity of protein-polymer interaction parameterization at the CG level to avoid overestimation of polymer association when employing other PEG models within the Martini framework.

Transferable high dimensional neural network potentials (HDNNP) have shown great promise as an avenue to increase the accuracy and domain of applicability of existing atomistic force fields for organic systems relevant to life science. We have previously reported such a potential (Schrödinger-ANI) that has broad coverage of druglike molecules. We extend that work here to cover ionic and zwitterionic druglike molecules expected to be relevant to drug discovery research activities. We report a novel HDNNP architechture, which we call QRNN, that predicts atomic charges and uses these charges as descriptors in an energy model which delivers conformational energies within chemical accuracy when measured against the reference theory it is trained to. Further, we find that delta learning based on a semi-empirical level of theory approximately halves the errors. We test the models on torsion energy profiles, relative conformational energies, geometric parameters and relative tautomer errors.

Liquid electrolytes are the most common electrolyte classes used in Li-ion batteries, which are a critical technology of the modern world. However, we still lack the computational tools required to accurately calculate key properties of these materials (viscosity, ionic diffusivity) from first principles necessary to sup-port improved designs. In this work, we report a machine learning-based force field for liquid electrolyte simulations which bridges the gap between the accuracy of range-separated hybrid density functional theory and the efficiency of classical force fields. Predictions of material properties made with this force field are quantitatively accurate compared to experimental data. Our model uses the QRNN deep neural network architecture, which includes both long-range interactions and global charge equilibration. The training dataset is composed solely of non-periodic DFT, allowing the practical use of an accurate theory (here, ωB97X-D3BJ/def2-TZVPD) which would be prohibitively expensive for generating large datasets with periodic DFT. In this report we focus on seven common carbonates and LiPF6, but this methodology has very few assumptions and can be readily applied to any liquid electrolyte system. This provides a promising path forward for large scale atomistic modeling of many important battery chemistries.