OrbNet: Deep Learning for Quantum Chemistry Using Symmetry-Adapted Atomic-Orbital Features

TL;DR

OrbNet is a deep learning approach that uses symmetry-adapted atomic orbital features and graph neural networks to accurately predict quantum chemistry energies with significantly reduced computational cost.

Contribution

This paper introduces OrbNet, a novel machine learning method combining symmetry-adapted features and GNNs for efficient quantum chemistry predictions.

Findings

Outperforms existing methods in learning efficiency and transferability.

Achieves chemical accuracy in energy predictions for diverse datasets.

Reduces computational cost by over a thousand times.

Abstract

We introduce a machine learning method in which energy solutions from the Schrodinger equation are predicted using symmetry adapted atomic orbitals features and a graph neural-network architecture. \textsc{OrbNet} is shown to outperform existing methods in terms of learning efficiency and transferability for the prediction of density functional theory results while employing low-cost features that are obtained from semi-empirical electronic structure calculations. For applications to datasets of drug-like molecules, including QM7b-T, QM9, GDB-13-T, DrugBank, and the conformer benchmark dataset of Folmsbee and Hutchison, \textsc{OrbNet} predicts energies within chemical accuracy of DFT at a computational cost that is thousand-fold or more reduced.