Learning a Local-Variable Model of Aromatic and Conjugated Systems
Open Access
- 3 January 2018
- journal article
- research article
- Published by American Chemical Society (ACS) in ACS Central Science
- Vol. 4 (1) , 52-62
Abstract
A collection of new approaches to building and training neural networks, collectively referred to as deep learning, are attracting attention in theoretical chemistry. Several groups aim to replace computationally expensive ab initio quantum mechanics calculations with learned estimators. This raises questions about the representability of complex quantum chemical systems with neural networks. Can local-variable models efficiently approximate nonlocal quantum chemical features? Here, we find that convolutional architectures, those that only aggregate information locally, cannot efficiently represent aromaticity and conjugation in large systems. They cannot represent long-range nonlocality known to be important in quantum chemistry. This study uses aromatic and conjugated systems computed from molecule graphs, though reproducing quantum simulations is the ultimate goal. This task, by definition, is both computable and known to be important to chemistry. The failure of convolutional architectures on this focused task calls into question their use in modeling quantum mechanics. To remedy this heretofore unrecognized deficiency, we introduce a new architecture that propagates information back and forth in waves of nonlinear computation. This architecture is still a local-variable model, and it is both computationally and representationally efficient, processing molecules in sublinear time with far fewer parameters than convolutional networks. Wave-like propagation models aromatic and conjugated systems with high accuracy, and even models the impact of small structural changes on large molecules. This new architecture demonstrates that some nonlocal features of quantum chemistry can be efficiently represented in local variable models.Keywords
Funding Information
- U.S. National Library of Medicine (R01LM012222, R01LM012482)
- National Institute of General Medical Sciences (GM07200)
- Center for Biological Systems Engineering, Washington University in St. Louis
- Department of Pathology and Immunology, Washington University School of Medicine in St. Louis
This publication has 47 references indexed in Scilit:
- Machine learning of molecular electronic properties in chemical compound spaceNew Journal of Physics, 2013
- Open Babel: An open chemical toolboxJournal of Cheminformatics, 2011
- Learning to Predict Chemical ReactionsJournal of Chemical Information and Modeling, 2011
- Walking and orbiting dropletsNature, 2005
- The Chemistry Development Kit (CDK): An Open-Source Java Library for Chemo- and BioinformaticsJournal of Chemical Information and Computer Sciences, 2003
- A History of the Structural Theory of Benzene - The Aromatic Sextet Rule and Huckel's RuleJournal of Chemical Education, 1997
- Properties of cross-entropy minimizationIEEE Transactions on Information Theory, 1981
- Aromaticity and conjugationJournal of the American Chemical Society, 1977
- Relationship between Absorption Spectrum and Molecular Conformations of 11-cis-RetinalNature, 1969
- A Suggested Interpretation of the Quantum Theory in Terms of "Hidden" Variables. IPhysical Review B, 1952