A shared-weight neural network architecture for predicting molecular properties

文献情報

出版日 2019-11-12

DOI 10.1039/C9CP03103K

インパクトファクター 3.676

著者

Trevor A. Profitt, Jason K. Pearson

原文を見る

要旨

Quantum chemical methods scale poorly with increasing molecular size and machine learning models have emerged as a promising, computationally-efficient alternative. We present a shared-weight neural network architecture based on modified atom-centered symmetry functions (ACSFs) and show that it performs similarly to the more computationally expensive per-element neural networks of previous work with ACSFs. The model achieves chemically accurate predictions, with a mean absolute error as low as 0.63 kcal mol−1 on energy predictions in the QM9 data set. Additionally, we show that it can reliably predict atomic forces.

掲載誌

Physical Chemistry Chemical Physics

CiteScore: 5.5

自己引用率: 10.3%

年間論文数: 3036

Physical Chemistry Chemical Physics (PCCP) is an international journal co-owned by 19 physical chemistry and physics societies from around the world. This journal publishes original, cutting-edge research in physical chemistry, chemical physics and biophysical chemistry. To be suitable for publication in PCCP, articles must include significant innovation and/or insight into physical chemistry; this is the most important criterion that reviewers and Editors will judge against when evaluating submissions. The journal has a broad scope and welcomes contributions spanning experiment, theory, computation and data science. Topical coverage includes spectroscopy, dynamics, kinetics, statistical mechanics, thermodynamics, electrochemistry, catalysis, surface science, quantum mechanics, quantum computing and machine learning. Interdisciplinary research areas such as polymers and soft matter, materials, nanoscience, energy, surfaces/interfaces, and biophysical chemistry are welcomed if they demonstrate significant innovation and/or insight into physical chemistry. Joined experimental/theoretical studies are particularly appreciated when complementary and based on up-to-date approaches.