Transferable Dynamic Molecular Charge Assignment Using Deep Neural Networks

doi:10.1021/acs.jctc.8b00524

Title: Transferable Dynamic Molecular Charge Assignment Using Deep Neural Networks

Abstract

We report that the ability to accurately and efficiently compute quantum-mechanical partial atomistic charges has many practical applications, such as calculations of IR spectra, analysis of chemical bonding, and classical force field parametrization. Machine learning (ML) techniques provide a possible avenue for the efficient prediction of atomic partial charges. Modern ML advances in the prediction of molecular energies [i.e., the hierarchical interacting particle neural network (HIP-NN)] has provided the necessary model framework and architecture to predict transferable, extensible, and conformationally dynamic atomic partial charges based on reference density functional theory (DFT) simulations. Utilizing HIP-NN, we show that ML charge prediction can be highly accurate over a wide range of molecules (both small and large) across a variety of charge partitioning schemes such as the Hirshfeld, CM5, MSK, and NBO methods. To demonstrate transferability and size extensibility, we compare ML results with reference DFT calculations on the COMP6 benchmark, achieving errors of 0.004e ^– (elementary charge). This is remarkable since this benchmark contains two proteins that are multiple times larger than the largest molecules in the training set. An application of our atomic charge predictions on nonequilibrium geometries is the generation of IR spectra for organic molecules from dynamical trajectories onmore »« less

Authors:

^[1];

^[2]; Sifain, Andrew E. ^[3];

^[1]; Isayev, Olexandr ^[4]; Roitberg, Adrian ^[5];

^[1];

^[1]

Los Alamos National Lab. (LANL), Los Alamos, NM (United States)
Los Alamos National Lab. (LANL), Los Alamos, NM (United States); Univ. of Florida, Gainesville, FL (United States)
Los Alamos National Lab. (LANL), Los Alamos, NM (United States); Univ. of Southern California, Los Angeles, CA (United States)
Univ. of North Carolina, Chapel Hill, NC (United States)
Univ. of Florida, Gainesville, FL (United States)

Publication Date:: 2018-07-31

Research Org.:: Los Alamos National Lab. (LANL), Los Alamos, NM (United States)

Sponsoring Org.:: USDOE National Nuclear Security Administration (NNSA); USDOE Laboratory Directed Research and Development (LDRD) Program

OSTI Identifier:: 1467336

Report Number(s):: LA-UR-18-22005
Journal ID: ISSN 1549-9618

Grant/Contract Number:: AC52-06NA25396

Resource Type:: Accepted Manuscript

Journal Name:: Journal of Chemical Theory and Computation

Additional Journal Information:: Journal Volume: 14; Journal Issue: 9; Journal ID: ISSN 1549-9618

Publisher:: American Chemical Society

Country of Publication:: United States

Language:: English

Subject:: 37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CHEMISTRY; 97 MATHEMATICS AND COMPUTING

Citation Formats


                    Nebgen, Benjamin Tyler, Lubbers, Nicholas Edward, Smith, Justin Steven, Sifain, Andrew E., Lokhov, Andrey, Isayev, Olexandr, Roitberg, Adrian, Barros, Kipton Marcos, and Tretiak, Sergei. Transferable Dynamic Molecular Charge Assignment Using Deep Neural Networks.  United States: N. p., 2018. 
        Web.  doi:10.1021/acs.jctc.8b00524.

Copy to clipboard


                    Nebgen, Benjamin Tyler, Lubbers, Nicholas Edward, Smith, Justin Steven, Sifain, Andrew E., Lokhov, Andrey, Isayev, Olexandr, Roitberg, Adrian, Barros, Kipton Marcos, & Tretiak, Sergei. Transferable Dynamic Molecular Charge Assignment Using Deep Neural Networks.  United States.  doi:10.1021/acs.jctc.8b00524.

Copy to clipboard


                    Nebgen, Benjamin Tyler, Lubbers, Nicholas Edward, Smith, Justin Steven, Sifain, Andrew E., Lokhov, Andrey, Isayev, Olexandr, Roitberg, Adrian, Barros, Kipton Marcos, and Tretiak, Sergei. Tue .  
        "Transferable Dynamic Molecular Charge Assignment Using Deep Neural Networks".  United States.  doi:10.1021/acs.jctc.8b00524.  https://www.osti.gov/servlets/purl/1467336.

Copy to clipboard


                    
@article{osti_1467336,

  title        = {Transferable Dynamic Molecular Charge Assignment Using Deep Neural Networks},

  author       = {Nebgen, Benjamin Tyler and Lubbers, Nicholas Edward and Smith, Justin Steven and Sifain, Andrew E. and Lokhov, Andrey and Isayev, Olexandr and Roitberg, Adrian and Barros, Kipton Marcos and Tretiak, Sergei},

  abstractNote = {We report that the ability to accurately and efficiently compute quantum-mechanical partial atomistic charges has many practical applications, such as calculations of IR spectra, analysis of chemical bonding, and classical force field parametrization. Machine learning (ML) techniques provide a possible avenue for the efficient prediction of atomic partial charges. Modern ML advances in the prediction of molecular energies [i.e., the hierarchical interacting particle neural network (HIP-NN)] has provided the necessary model framework and architecture to predict transferable, extensible, and conformationally dynamic atomic partial charges based on reference density functional theory (DFT) simulations. Utilizing HIP-NN, we show that ML charge prediction can be highly accurate over a wide range of molecules (both small and large) across a variety of charge partitioning schemes such as the Hirshfeld, CM5, MSK, and NBO methods. To demonstrate transferability and size extensibility, we compare ML results with reference DFT calculations on the COMP6 benchmark, achieving errors of 0.004e– (elementary charge). This is remarkable since this benchmark contains two proteins that are multiple times larger than the largest molecules in the training set. An application of our atomic charge predictions on nonequilibrium geometries is the generation of IR spectra for organic molecules from dynamical trajectories on a variety of organic molecules, which show good agreement with calculated IR spectra with reference method. Critically, HIP-NN charge predictions are many orders of magnitude faster than direct DFT calculations. Lastly, these combined results provide further evidence that ML (specifically HIP-NN) provides a pathway to greatly increase the range of feasible simulations while retaining quantum-level accuracy.},

  doi          = {10.1021/acs.jctc.8b00524},

  journal      = {Journal of Chemical Theory and Computation},

  number       = 9,

  volume       = 14,

  place        = {United States},

  year         = {2018},

  month        = {7}

}

Copy to clipboard

Journal Article:

Free Publicly Available Full Text

Accepted Manuscript (DOE)

Publisher's Version of Record

DOI: 10.1021/acs.jctc.8b00524

Other availability

Search WorldCat to find libraries that may hold this journal

Citation Metrics:

Cited by: 10 works

Citation information provided by
Web of Science

Save / Share:

Export Metadata

Save to My Library

Works referencing / citing this record:

Approaching coupled cluster accuracy with a general-purpose neural network potential through transfer learning
journal, July 2019

Smith, Justin S.; Nebgen, Benjamin T.; Zubatyuk, Roman
Nature Communications, Vol. 10, Issue 1
DOI: 10.1038/s41467-019-10827-4

Similar records in OSTI.GOV collections:

Modeling the Partial Atomic Charges in Inorganometallic Molecules and Solids and Charge Redistribution in Lithium-Ion Cathodes

Journal Article Wang, Bo ; Li, Shaohong L. ; Truhlar, Donald G. - Journal of Chemical Theory and Computation

Partial atomic charges are widely used for the description of charge distributions of molecules and solids. These charges are useful to indicate the extent of charge transfer and charge flow during chemical reactions in batteries, fuel cells, and catalysts and to characterize charge distributions in capacitors, liquid-phase electrolytes, and solids and at electrochemical interfaces. However, partial atomic charges given by various charge models differ significantly, especially for systems containing metal atoms. In the present study, we have compared various charge models on both molecular systems and extended systems, including Hirshfeld, CM5, MK, ChElPG, Mulliken, MBS, NPA, DDEC, LoProp, and Badermore »« less
Cited by 14
DOI: 10.1021/ct500790p

Full Text Available
Modeling the Partial Atomic Charges in Inorganometallic Molecules and Solids and Charge Redistribution in Lithium-Ion Cathodes

Journal Article Wang, Bo ; Li, Shaohong L. ; Truhlar, Donald G. - Journal of Chemical Theory and Computation

Partial atomic charges are widely used for the description of charge distributions of molecules and solids. These charges are useful to indicate the extent of charge transfer and charge flow during chemical reactions in batteries, fuel cells, and catalysts and to characterize charge distributions in capacitors, liquid-phase electrolytes, and solids and at electrochemical interfaces. However, partial atomic charges given by various charge models differ significantly, especially for systems containing metal atoms. In the present study, we have compared various charge models on both molecular systems and extended systems, including Hirshfeld, CM5, MK, ChElPG, Mulliken, MBS, NPA, DDEC, LoProp, and Badermore »« less
Cited by 14
DOI: 10.1021/ct500790p
Less is more: Sampling chemical space with active learning

Journal Article Smith, Justin Steven ; Nebgen, Benjamin Tyler ; Lubbers, Nicholas Edward ; ... - Journal of Chemical Physics

We present the development of accurate and transferable machine learning (ML) potentials for predicting molecular energetics is a challenging task. The process of data generation to train such ML potentials is a task neither well understood nor researched in detail. In this work, we present a fully automated approach for the generation of datasets with the intent of training universal ML potentials. It is based on the concept of active learning (AL) via Query by Committee (QBC), which uses the disagreement between an ensemble of ML potentials to infer the reliability of the ensemble’s prediction. QBC allows the presented ALmore »« less
Cited by 19
DOI: 10.1063/1.5023802

Full Text Available
Numerical study on the partitioning of the molecular polarizability into fluctuating charge and induced atomic dipole contributions

Journal Article Mei, Ye ; Simmonett, Andrew C. ; Pickard, IV, Frank C. ; ... - Journal of Physical Chemistry. A, Molecules, Spectroscopy, Kinetics, Environment, and General Theory

In order to carry out a detailed analysis of the molecular static polarizability, which is the response of the molecule to a uniform external electric field, the molecular polarizability was computed in this study using the finite-difference method for 21 small molecules, using density functional theory. Within nine charge population schemes (Lowdin, Mulliken, Becke, Hirshfeld, CM5, Hirshfeld-I, NPA, CHELPG, MK-ESP) in common use, the charge fluctuation contribution is found to dominate the molecular polarizability, with its ratio ranging from 59.9% with the Hirshfeld or CM5 scheme to 96.2% with the Mulliken scheme. The Hirshfeld-I scheme is also used to computemore »« less
Cited by 7
DOI: 10.1021/acs.jpca.5b03159

Full Text Available
A Simple AIMD Approach to Derive Atomic Charges for Condensed Phase Simulation of Ionic Liquids

Journal Article Zhang, Y ; Maginn, EJ - Journal of Physical Chemistry. B, Condensed Matter, Materials, Surfaces, Interfaces and Biophysical Chemistry

The atomic charges for two ionic liquids (ILs), 1-n-butyl-3-methylimidazolium hexafluorophosphate ([BMIM] [PF6]) and 1-ethyl-3-methylimidazolium hexafluorophosphate ([EMIM][PF6]), were derived from periodic crystal phase calculations with density functional theory (DFT) and plane wave basis sets (denoted as "AIMD-c charge"). For both ILs, the total charge was found to be +/- 0.8 e for the cation and anion, respectively, due to the charge transfer between ions and polarization caused by the environment. These atomic charges were used in a force field developed within the general Amber force field framework. Using this force field, static, dynamic, and thermodynamic properties were computed for the twomore »« less
DOI: 10.1021/jp3037999

Similar Records

Title: Transferable Dynamic Molecular Charge Assignment Using Deep Neural Networks

Abstract

Citation Formats

Approaching coupled cluster accuracy with a general-purpose neural network potential through transfer learning journal, July 2019

Approaching coupled cluster accuracy with a general-purpose neural network potential through transfer learning
journal, July 2019