DOE PAGES title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Accurate and transferable multitask prediction of chemical properties with an atoms-in-molecules neural network

Abstract

Atomic and molecular properties could be evaluated from the fundamental Schrodinger’s equation and therefore represent different modalities of the same quantum phenomena. Here, we present AIMNet, a modular and chemically inspired deep neural network potential. We used AIMNet with multitarget training to learn multiple modalities of the state of the atom in a molecular system. The resulting model shows on several benchmark datasets state-of-the-art accuracy, comparable to the results of orders of magnitude more expensive DFT methods. It can simultaneously predict several atomic and molecular properties without an increase in the computational cost. With AIMNet, we show a new dimension of transferability: the ability to learn new targets using multimodal information from previous training. The model can learn implicit solvation energy (SMD method) using only a fraction of the original training data and an archive median absolute deviation error of 1.1 kcal/mol compared to experimental solvation free energies in the MNSol database.

Authors:
ORCiD logo [1]; ORCiD logo [2]; ORCiD logo [3]; ORCiD logo [4]
  1. Univ. of North Carolina, Chapel Hill, NC (United States); Los Alamos National Lab. (LANL), Los Alamos, NM (United States); Jackson State Univ., Jackson, MS (United States)
  2. Los Alamos National Lab. (LANL), Los Alamos, NM (United States)
  3. Jackson State Univ., Jackson, MS (United States)
  4. Univ. of North Carolina, Chapel Hill, NC (United States)
Publication Date:
Research Org.:
Los Alamos National Lab. (LANL), Los Alamos, NM (United States)
Sponsoring Org.:
USDOE National Nuclear Security Administration (NNSA)
OSTI Identifier:
1570635
Report Number(s):
LA-UR-18-29299
Journal ID: ISSN 2375-2548
Grant/Contract Number:  
89233218CNA000001
Resource Type:
Accepted Manuscript
Journal Name:
Science Advances
Additional Journal Information:
Journal Volume: 5; Journal Issue: 8; Journal ID: ISSN 2375-2548
Publisher:
AAAS
Country of Publication:
United States
Language:
English
Subject:
37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CHEMISTRY

Citation Formats

Zubatyuk, Roman, Smith, Justin Steven, Leszczynski, Jerzy, and Isayev, Olexandr. Accurate and transferable multitask prediction of chemical properties with an atoms-in-molecules neural network. United States: N. p., 2019. Web. doi:10.1126/sciadv.aav6490.
Zubatyuk, Roman, Smith, Justin Steven, Leszczynski, Jerzy, & Isayev, Olexandr. Accurate and transferable multitask prediction of chemical properties with an atoms-in-molecules neural network. United States. https://doi.org/10.1126/sciadv.aav6490
Zubatyuk, Roman, Smith, Justin Steven, Leszczynski, Jerzy, and Isayev, Olexandr. Fri . "Accurate and transferable multitask prediction of chemical properties with an atoms-in-molecules neural network". United States. https://doi.org/10.1126/sciadv.aav6490. https://www.osti.gov/servlets/purl/1570635.
@article{osti_1570635,
title = {Accurate and transferable multitask prediction of chemical properties with an atoms-in-molecules neural network},
author = {Zubatyuk, Roman and Smith, Justin Steven and Leszczynski, Jerzy and Isayev, Olexandr},
abstractNote = {Atomic and molecular properties could be evaluated from the fundamental Schrodinger’s equation and therefore represent different modalities of the same quantum phenomena. Here, we present AIMNet, a modular and chemically inspired deep neural network potential. We used AIMNet with multitarget training to learn multiple modalities of the state of the atom in a molecular system. The resulting model shows on several benchmark datasets state-of-the-art accuracy, comparable to the results of orders of magnitude more expensive DFT methods. It can simultaneously predict several atomic and molecular properties without an increase in the computational cost. With AIMNet, we show a new dimension of transferability: the ability to learn new targets using multimodal information from previous training. The model can learn implicit solvation energy (SMD method) using only a fraction of the original training data and an archive median absolute deviation error of 1.1 kcal/mol compared to experimental solvation free energies in the MNSol database.},
doi = {10.1126/sciadv.aav6490},
journal = {Science Advances},
number = 8,
volume = 5,
place = {United States},
year = {2019},
month = {8}
}

Journal Article:
Free Publicly Available Full Text
Publisher's Version of Record

Citation Metrics:
Cited by: 15 works
Citation information provided by
Web of Science

Figures / Tables:

Fig. 1 Fig. 1: Architecture of the AIMNet model. The model uses atomic numbers Z and coordinates R as input features. The coordinates are transformed with ANI-type symmetry functions into AEVs. Atom types are represented with learnable atomic feature vectors (AFV), which are used as embedding vectors for AEVs. The interaction ofmore » an atom with its environment produces the AIM representation of the atom used to predict a set of target atom properties {pk}. The environment-dependent update to AFV within N iterations is used to make the embedding vectors for each atom consistent with its environment. Input data are colored in blue, predicted endpoints are in orange, and neural network blocks are in green.« less

Save / Share:

Works referenced in this record:

Fast and Accurate Modeling of Molecular Atomization Energies with Machine Learning
journal, January 2012


The Pilot Way to Grid Resources Using glideinWMS
conference, March 2009

  • Sfiligoi, Igor; Bradley, Daniel C.; Holzman, Burt
  • 2009 WRI World Congress on Computer Science and Information Engineering
  • DOI: 10.1109/CSIE.2009.950

The TensorMol-0.1 model chemistry: a neural network augmented with long-range physics
journal, January 2018

  • Yao, Kun; Herr, John E.; Toth, David W.
  • Chemical Science, Vol. 9, Issue 8
  • DOI: 10.1039/C7SC04934J

Inverse molecular design using machine learning: Generative models for matter engineering
journal, July 2018


ANI-1: an extensible neural network potential with DFT accuracy at force field computational cost
journal, January 2017

  • Smith, J. S.; Isayev, O.; Roitberg, A. E.
  • Chemical Science, Vol. 8, Issue 4
  • DOI: 10.1039/C6SC05720A

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


Improving solvation energy predictions using the SMD solvation method and semiempirical electronic structure methods
journal, September 2018

  • Kromann, Jimmy C.; Steinmann, Casper; Jensen, Jan H.
  • The Journal of Chemical Physics, Vol. 149, Issue 10
  • DOI: 10.1063/1.5047273

The ORCA program system: The ORCA program system
journal, June 2011

  • Neese, Frank
  • Wiley Interdisciplinary Reviews: Computational Molecular Science, Vol. 2, Issue 1
  • DOI: 10.1002/wcms.81

Less is more: Sampling chemical space with active learning
journal, June 2018

  • Smith, Justin S.; Nebgen, Ben; Lubbers, Nicholas
  • The Journal of Chemical Physics, Vol. 148, Issue 24
  • DOI: 10.1063/1.5023802

Predicting electronic structure properties of transition metal complexes with neural networks
journal, January 2017

  • Janet, Jon Paul; Kulik, Heather J.
  • Chemical Science, Vol. 8, Issue 7
  • DOI: 10.1039/C7SC01247K

Generalized Neural-Network Representation of High-Dimensional Potential-Energy Surfaces
journal, April 2007


The S66x8 benchmark for noncovalent interactions revisited: explicitly correlated ab initio methods and density functional theory
journal, January 2016

  • Brauer, Brina; Kesharwani, Manoj K.; Kozuch, Sebastian
  • Physical Chemistry Chemical Physics, Vol. 18, Issue 31
  • DOI: 10.1039/C6CP00688D

Toward Accurate Conformational Energies of Smaller Peptides and Medium-Sized Macrocycles: MPCONF196 Benchmark Energy Data Set
journal, February 2018

  • Řezáč, Jan; Bím, Daniel; Gutten, Ondrej
  • Journal of Chemical Theory and Computation, Vol. 14, Issue 3
  • DOI: 10.1021/acs.jctc.7b01074

Brain mechanisms linking language and action
journal, June 2005

  • Pulvermüller, Friedemann
  • Nature Reviews Neuroscience, Vol. 6, Issue 7
  • DOI: 10.1038/nrn1706

A Density Functional Tight Binding Layer for Deep Learning of Chemical Hamiltonians
journal, October 2018

  • Li, Haichen; Collins, Christopher; Tanha, Matteus
  • Journal of Chemical Theory and Computation, Vol. 14, Issue 11
  • DOI: 10.1021/acs.jctc.8b00873

A Comparison of Quantum and Molecular Mechanical Methods to Estimate Strain Energy in Druglike Fragments
journal, May 2017

  • Sellers, Benjamin D.; James, Natalie C.; Gobbi, Alberto
  • Journal of Chemical Information and Modeling, Vol. 57, Issue 6
  • DOI: 10.1021/acs.jcim.6b00614

Transferability in Machine Learning for Electronic Structure via the Molecular Orbital Basis
journal, July 2018

  • Welborn, Matthew; Cheng, Lixue; Miller, Thomas F.
  • Journal of Chemical Theory and Computation, Vol. 14, Issue 9
  • DOI: 10.1021/acs.jctc.8b00636

SchNet – A deep learning architecture for molecules and materials
journal, June 2018

  • Schütt, K. T.; Sauceda, H. E.; Kindermans, P. -J.
  • The Journal of Chemical Physics, Vol. 148, Issue 24
  • DOI: 10.1063/1.5019779

Exploring the use of adaptive gradient methods in effective deep learning systems
conference, April 2018


The open science grid
journal, July 2007


Machine learning molecular dynamics for the simulation of infrared spectra
journal, January 2017

  • Gastegger, Michael; Behler, Jörg; Marquetand, Philipp
  • Chemical Science, Vol. 8, Issue 10
  • DOI: 10.1039/C7SC02267K

Accelerating Chemical Discovery with Machine Learning: Simulated Evolution of Spin Crossover Complexes with an Artificial Neural Network
journal, February 2018

  • Janet, Jon Paul; Chan, Lydia; Kulik, Heather J.
  • The Journal of Physical Chemistry Letters, Vol. 9, Issue 5
  • DOI: 10.1021/acs.jpclett.8b00170

Minimal Basis Iterative Stockholder: Atoms in Molecules for Force-Field Development
journal, July 2016

  • Verstraelen, Toon; Vandenbrande, Steven; Heidar-Zadeh, Farnaz
  • Journal of Chemical Theory and Computation, Vol. 12, Issue 8
  • DOI: 10.1021/acs.jctc.6b00456

Machine learning of accurate energy-conserving molecular force fields
journal, May 2017

  • Chmiela, Stefan; Tkatchenko, Alexandre; Sauceda, Huziel E.
  • Science Advances, Vol. 3, Issue 5
  • DOI: 10.1126/sciadv.1603015

First Principles Neural Network Potentials for Reactive Simulations of Large Molecular and Condensed Systems
journal, August 2017


Machine learning for molecular and materials science
journal, July 2018


Planning chemical syntheses with deep neural networks and symbolic AI
journal, March 2018

  • Segler, Marwin H. S.; Preuss, Mike; Waller, Mark P.
  • Nature, Vol. 555, Issue 7698
  • DOI: 10.1038/nature25978

An overlap fitted chain of spheres exchange method
journal, October 2011

  • Izsák, Róbert; Neese, Frank
  • The Journal of Chemical Physics, Vol. 135, Issue 14
  • DOI: 10.1063/1.3646921

Active learning of linearly parametrized interatomic potentials
journal, December 2017


Discovering a Transferable Charge Assignment Model Using Machine Learning
journal, July 2018

  • Sifain, Andrew E.; Lubbers, Nicholas; Nebgen, Benjamin T.
  • The Journal of Physical Chemistry Letters, Vol. 9, Issue 16
  • DOI: 10.1021/acs.jpclett.8b01939

RPLS-Net: pulmonary lobe segmentation based on 3D fully convolutional networks and multi-task learning
journal, April 2021

  • Liu, Jinxin; Wang, Chengdi; Guo, Jixiang
  • International Journal of Computer Assisted Radiology and Surgery, Vol. 16, Issue 6
  • DOI: 10.1007/s11548-021-02360-x

ANI-1: an extensible neural network potential with DFT accuracy at force field computational cost
text, January 2017

  • ,
  • The University of North Carolina at Chapel Hill University Libraries
  • DOI: 10.17615/bhbf-9r93

Imitation of Life: Structure, Agency and Discourse in Theatrical Performance
journal, July 2012


Fast and accurate modeling of molecular atomization energies with machine learning
text, January 2012

  • Rupp, Matthias; Tkatchenko, Alexandre; Müller, Klaus-Robert
  • American Physical Society
  • DOI: 10.5451/unibas-ep43360

Figures/Tables have been extracted from DOE-funded journal article accepted manuscripts.