OrbNet：利用对称适配原子轨道特征进行量子化学的深度学习

OrbNet: Deep learning for quantum chemistry using symmetry-adapted atomic-orbital features.

作者信息

Qiao Zhuoran, Welborn Matthew, Anandkumar Animashree, Manby Frederick R, Miller Thomas F

机构信息

Division of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena, California 91125, USA.

Entos, Inc., 4470 W Sunset Blvd., Suite 107 PMB 94758, Los Angeles, California 90027, USA.

出版信息

J Chem Phys. 2020 Sep 28;153(12):124111. doi: 10.1063/5.0021955.

DOI:10.1063/5.0021955

PMID:33003742

Abstract

We introduce a machine learning method in which energy solutions from the Schrödinger equation are predicted using symmetry adapted atomic orbital features and a graph neural-network architecture. 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 [Int. J. Quantum Chem. (published online) (2020)], OrbNet predicts energies within chemical accuracy of density functional theory at a computational cost that is 1000-fold or more reduced.

摘要

我们介绍了一种机器学习方法，其中利用对称适配原子轨道特征和图神经网络架构来预测薛定谔方程的能量解。结果表明，在预测密度泛函理论结果时，OrbNet在学习效率和可迁移性方面优于现有方法，同时采用了从半经验电子结构计算中获得的低成本特征。对于应用于类药物分子数据集，包括QM7b-T、QM9、GDB-13-T、DrugBank以及Folmsbee和Hutchison的构象异构体基准数据集[《国际量子化学杂志》（在线发表）（2020年）]，OrbNet能够以降低1000倍或更多的计算成本预测出在密度泛函理论化学精度范围内的能量。