Analytical gradients for molecular-orbital-based machine learning
Abstract
We report molecular-orbital-based machine learning (MOB-ML) enables the prediction of accurate correlation energies at the cost of obtaining molecular orbitals. Here, we present the derivation, implementation, and numerical demonstration of MOB-ML analytical nuclear gradients, which are formulated in a general Lagrangian framework to enforce orthogonality, localization, and Brillouin constraints on the molecular orbitals. The MOB-ML gradient framework is general with respect to the regression technique (e.g., Gaussian process regression or neural networks) and the MOB feature design. We show that MOB-ML gradients are highly accurate compared to other ML methods on the ISO17 dataset while only being trained on energies for hundreds of molecules compared to energies and gradients for hundreds of thousands of molecules for the other ML methods. The MOB-ML gradients are also shown to yield accurate optimized structures at a computational cost for the gradient evaluation that is comparable to a density-corrected density functional theory calculation.
- Authors:
-
- California Institute of Technology (CalTech), Pasadena, CA (United States)
- Entos, Inc., Los Angeles, CA (United States)
- California Institute of Technology (CalTech), Pasadena, CA (United States); Entos, Inc., Los Angeles, CA (United States)
- Publication Date:
- Research Org.:
- California Institute of Technology (CalTech), Pasadena, CA (United States); Univ. of California, Oakland, CA (United States)
- Sponsoring Org.:
- USDOE Office of Science (SC); U.S. Army Research Laboratory; Caltech DeLogi Fund; Camille and Henry Dreyfus Foundation; Swiss National Science Foundation (SNSF)
- OSTI Identifier:
- 1853179
- Alternate Identifier(s):
- OSTI ID: 1988100
- Grant/Contract Number:
- SC0019390; AC02-05CH11231; W911NF-12-2-0023; ML-20-196; P2EZP2_184234
- Resource Type:
- Accepted Manuscript
- Journal Name:
- Journal of Chemical Physics
- Additional Journal Information:
- Journal Volume: 154; Journal Issue: 12; Journal ID: ISSN 0021-9606
- Publisher:
- American Institute of Physics (AIP)
- Country of Publication:
- United States
- Language:
- English
- Subject:
- 71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSICS; 37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CHEMISTRY; energy forecasting; Gaussian processes; isomerism; correlation-consistent basis sets; regression analysis; artificial neural networks; machine learning; mathematical optimization; correlation energy; coupled-cluster methods
Citation Formats
Lee, Sebastian R., Husch, Tamara, Ding, Feizhi, and Miller, Thomas F. Analytical gradients for molecular-orbital-based machine learning. United States: N. p., 2021.
Web. doi:10.1063/5.0040782.
Lee, Sebastian R., Husch, Tamara, Ding, Feizhi, & Miller, Thomas F. Analytical gradients for molecular-orbital-based machine learning. United States. https://doi.org/10.1063/5.0040782
Lee, Sebastian R., Husch, Tamara, Ding, Feizhi, and Miller, Thomas F. Thu .
"Analytical gradients for molecular-orbital-based machine learning". United States. https://doi.org/10.1063/5.0040782. https://www.osti.gov/servlets/purl/1853179.
@article{osti_1853179,
title = {Analytical gradients for molecular-orbital-based machine learning},
author = {Lee, Sebastian R. and Husch, Tamara and Ding, Feizhi and Miller, Thomas F.},
abstractNote = {We report molecular-orbital-based machine learning (MOB-ML) enables the prediction of accurate correlation energies at the cost of obtaining molecular orbitals. Here, we present the derivation, implementation, and numerical demonstration of MOB-ML analytical nuclear gradients, which are formulated in a general Lagrangian framework to enforce orthogonality, localization, and Brillouin constraints on the molecular orbitals. The MOB-ML gradient framework is general with respect to the regression technique (e.g., Gaussian process regression or neural networks) and the MOB feature design. We show that MOB-ML gradients are highly accurate compared to other ML methods on the ISO17 dataset while only being trained on energies for hundreds of molecules compared to energies and gradients for hundreds of thousands of molecules for the other ML methods. The MOB-ML gradients are also shown to yield accurate optimized structures at a computational cost for the gradient evaluation that is comparable to a density-corrected density functional theory calculation.},
doi = {10.1063/5.0040782},
journal = {Journal of Chemical Physics},
number = 12,
volume = 154,
place = {United States},
year = {Thu Mar 25 00:00:00 EDT 2021},
month = {Thu Mar 25 00:00:00 EDT 2021}
}
Works referenced in this record:
Neural Networks for the Prediction of Organic Chemistry Reactions
journal, October 2016
- Wei, Jennifer N.; Duvenaud, David; Aspuru-Guzik, Alán
- ACS Central Science, Vol. 2, Issue 10
Analytic gradients for coupled‐cluster energies that include noniterative connected triple excitations: Application to c i s ‐ and t r a n s ‐HONO
journal, May 1991
- Lee, Timothy J.; Rendell, Alistair P.
- The Journal of Chemical Physics, Vol. 94, Issue 9
Low-order scaling local electron correlation methods. I. Linear scaling local MP2
journal, October 1999
- Schütz, Martin; Hetzer, Georg; Werner, Hans-Joachim
- The Journal of Chemical Physics, Vol. 111, Issue 13
The closed‐shell coupled cluster single and double excitation (CCSD) model for the description of electron correlation. A comparison with configuration interaction (CISD) results
journal, March 1987
- Scuseria, Gustavo E.; Scheiner, Andrew C.; Lee, Timothy J.
- The Journal of Chemical Physics, Vol. 86, Issue 5
Towards exact molecular dynamics simulations with machine-learned force fields
journal, September 2018
- Chmiela, Stefan; Sauceda, Huziel E.; Müller, Klaus-Robert
- Nature Communications, Vol. 9, Issue 1
Deep reinforcement learning for de novo drug design
journal, July 2018
- Popova, Mariya; Isayev, Olexandr; Tropsha, Alexander
- Science Advances, Vol. 4, Issue 7
Analytical gradient for the domain-based local pair natural orbital second order Møller-Plesset perturbation theory method (DLPNO-MP2)
journal, April 2019
- Pinski, Peter; Neese, Frank
- The Journal of Chemical Physics, Vol. 150, Issue 16
An efficient local coupled cluster method for accurate thermochemistry of large systems
journal, October 2011
- Werner, Hans-Joachim; Schütz, Martin
- The Journal of Chemical Physics, Vol. 135, Issue 14
Analytical energy gradients for local second-order Møller–Plesset perturbation theory using density fitting approximations
journal, July 2004
- Schütz, Martin; Werner, Hans-Joachim; Lindh, Roland
- The Journal of Chemical Physics, Vol. 121, Issue 2
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
Gaussian basis sets for use in correlated molecular calculations. I. The atoms boron through neon and hydrogen
journal, January 1989
- Dunning, Thom H.
- The Journal of Chemical Physics, Vol. 90, Issue 2
Fast linear scaling second-order Møller-Plesset perturbation theory (MP2) using local and density fitting approximations
journal, May 2003
- Werner, Hans-Joachim; Manby, Frederick R.; Knowles, Peter J.
- The Journal of Chemical Physics, Vol. 118, Issue 18
Statistical Modeling: The Two Cultures (with comments and a rejoinder by the author)
journal, August 2001
- Breiman, Leo
- Statistical Science, Vol. 16, Issue 3
Electron affinities of the first‐row atoms revisited. Systematic basis sets and wave functions
journal, May 1992
- Kendall, Rick A.; Dunning, Thom H.; Harrison, Robert J.
- The Journal of Chemical Physics, Vol. 96, Issue 9
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
Analytic evaluation of energy gradients for the single and double excitation coupled cluster (CCSD) wave function: Theory and application
journal, November 1987
- Scheiner, Andrew C.; Scuseria, Gustavo E.; Rice, Julia E.
- The Journal of Chemical Physics, Vol. 87, Issue 9
Canonical Configurational Interaction Procedure
journal, April 1960
- Foster, J. M.; Boys, S. F.
- Reviews of Modern Physics, Vol. 32, Issue 2
Thermalized (350K) QM7b, GDB-13, water, and short alkane quantum chemistry dataset including MOB-ML features
February 2019
- Cheng, Lixue; Welborn, Matthew; Christensen, Anders S.
PhysNet: A Neural Network for Predicting Energies, Forces, Dipole Moments, and Partial Charges
journal, April 2019
- Unke, Oliver T.; Meuwly, Markus
- Journal of Chemical Theory and Computation, Vol. 15, Issue 6
SchNet: A continuous-filter convolutional neural network for modeling quantum interactions
text, January 2017
- Schütt, Kristof T.; Kindermans, Pieter-Jan; Sauceda, Huziel E.
- arXiv
A shared-weight neural network architecture for predicting molecular properties
journal, January 2019
- Profitt, Trevor A.; Pearson, Jason K.
- Physical Chemistry Chemical Physics, Vol. 21, Issue 47
Analytical gradients for projection-based wavefunction-in-DFT embedding
journal, August 2019
- Lee, Sebastian J. R.; Ding, Feizhi; Manby, Frederick R.
- The Journal of Chemical Physics, Vol. 151, Issue 6
Gaussian Moments as Physically Inspired Molecular Descriptors for Accurate and Scalable Machine Learning Potentials
journal, July 2020
- Zaverkin, V.; Kästner, J.
- Journal of Chemical Theory and Computation, Vol. 16, Issue 8
Deep Learning in Drug Discovery
journal, December 2015
- Gawehn, Erik; Hiss, Jan A.; Schneider, Gisbert
- Molecular Informatics, Vol. 35, Issue 1
Hierarchical modeling of molecular energies using a deep neural network
journal, June 2018
- Lubbers, Nicholas; Smith, Justin S.; Barros, Kipton
- The Journal of Chemical Physics, Vol. 148, Issue 24
Neural-Symbolic Machine Learning for Retrosynthesis and Reaction Prediction
journal, February 2017
- Segler, Marwin H. S.; Waller, Mark P.
- Chemistry - A European Journal, Vol. 23, Issue 25
Passing the one-billion limit in full configuration-interaction (FCI) calculations
journal, June 1990
- Olsen, Jeppe; Jørgensen, Poul; Simons, Jack
- Chemical Physics Letters, Vol. 169, Issue 6
Efficient use of the correlation consistent basis sets in resolution of the identity MP2 calculations
journal, February 2002
- Weigend, Florian; Köhn, Andreas; Hättig, Christof
- The Journal of Chemical Physics, Vol. 116, Issue 8
Optimization of the Linear-Scaling Local Natural Orbital CCSD(T) Method: Improved Algorithm and Benchmark Applications
journal, June 2018
- Nagy, Péter R.; Samu, Gyula; Kállay, Mihály
- Journal of Chemical Theory and Computation, Vol. 14, Issue 8
FCHL revisited: Faster and more accurate quantum machine learning
journal, January 2020
- Christensen, Anders S.; Bratholm, Lars A.; Faber, Felix A.
- The Journal of Chemical Physics, Vol. 152, Issue 4
Intrinsic Atomic Orbitals: An Unbiased Bridge between Quantum Theory and Chemical Concepts
journal, October 2013
- Knizia, Gerald
- Journal of Chemical Theory and Computation, Vol. 9, Issue 11
Improved accuracy and transferability of molecular-orbital-based machine learning: Organics, transition-metal complexes, non-covalent interactions, and transition states
journal, February 2021
- Husch, Tamara; Sun, Jiace; Cheng, Lixue
- The Journal of Chemical Physics, Vol. 154, Issue 6
A fully direct RI-HF algorithm: Implementation, optimised auxiliary basis sets, demonstration of accuracy and efficiency
journal, August 2002
- Weigend, Florian
- Physical Chemistry Chemical Physics, Vol. 4, Issue 18
Communication: An improved linear scaling perturbative triples correction for the domain based local pair-natural orbital based singles and doubles coupled cluster method [DLPNO-CCSD(T)]
journal, January 2018
- Guo, Yang; Riplinger, Christoph; Becker, Ute
- The Journal of Chemical Physics, Vol. 148, Issue 1
Analytical energy gradients for local second-order Møller-Plesset perturbation theory using intrinsic bond orbitals
journal, October 2018
- Dornbach, Mark; Werner, Hans-Joachim
- Molecular Physics, Vol. 117, Issue 9-12
Scalable Electron Correlation Methods. 3. Efficient and Accurate Parallel Local Coupled Cluster with Pair Natural Orbitals (PNO-LCCSD)
journal, July 2017
- Schwilk, Max; Ma, Qianli; Köppl, Christoph
- Journal of Chemical Theory and Computation, Vol. 13, Issue 8
Comparison of coupled‐cluster methods which include the effects of connected triple excitations
journal, October 1990
- Scuseria, Gustavo E.; Lee, Timothy J.
- The Journal of Chemical Physics, Vol. 93, Issue 8
Ab initio calculation of force constants and equilibrium geometries in polyatomic molecules : I. Theory
journal, January 1969
- Pulay, P.
- Molecular Physics, Vol. 17, Issue 2
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
A universal density matrix functional from molecular orbital-based machine learning: Transferability across organic molecules
journal, April 2019
- Cheng, Lixue; Welborn, Matthew; Christensen, Anders S.
- The Journal of Chemical Physics, Vol. 150, Issue 13
Virtual screening of inorganic materials synthesis parameters with deep learning
journal, December 2017
- Kim, Edward; Huang, Kevin; Jegelka, Stefanie
- npj Computational Materials, Vol. 3, Issue 1
Increasing the applicability of DFT I: Non-variational correlation corrections from Hartree–Fock DFT for predicting transition states
journal, February 2012
- Verma, Prakash; Perera, Ajith; Bartlett, Rodney J.
- Chemical Physics Letters, Vol. 524
Regression Clustering for Improved Accuracy and Training Costs with Molecular-Orbital-Based Machine Learning
journal, October 2019
- Cheng, Lixue; Kovachki, Nikola B.; Welborn, Matthew
- Journal of Chemical Theory and Computation, Vol. 15, Issue 12
Low-order scaling local correlation methods II: Splitting the Coulomb operator in linear scaling local second-order Møller–Plesset perturbation theory
journal, December 2000
- Hetzer, Georg; Schütz, Martin; Stoll, Hermann
- The Journal of Chemical Physics, Vol. 113, Issue 21
Low-order scaling local electron correlation methods. III. Linear scaling local perturbative triples correction ( T )
journal, December 2000
- Schütz, Martin
- The Journal of Chemical Physics, Vol. 113, Issue 22
Machine-learning approaches in drug discovery: methods and applications
journal, March 2015
- Lavecchia, Antonio
- Drug Discovery Today, Vol. 20, Issue 3
Machine-learning-assisted materials discovery using failed experiments
journal, May 2016
- Raccuglia, Paul; Elbert, Katherine C.; Adler, Philip D. F.
- Nature, Vol. 533, Issue 7601
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
To address surface reaction network complexity using scaling relations machine learning and DFT calculations
journal, March 2017
- Ulissi, Zachary W.; Medford, Andrew J.; Bligaard, Thomas
- Nature Communications, Vol. 8, Issue 1
Machine learning for molecular and materials science
journal, July 2018
- Butler, Keith T.; Davies, Daniel W.; Cartwright, Hugh
- Nature, Vol. 559, Issue 7715
Operators in quantum machine learning: Response properties in chemical space
journal, February 2019
- Christensen, Anders S.; Faber, Felix A.; von Lilienfeld, O. Anatole
- The Journal of Chemical Physics, Vol. 150, Issue 6
Accurate and scalable multi-element graph neural network force field and molecular dynamics with direct force architecture
preprint, January 2020
- Park, Cheol Woo; Kornbluth, Mordechai; Vandermause, Jonathan
- arXiv
Accurate and scalable graph neural network force field and molecular dynamics with direct force architecture
journal, May 2021
- Park, Cheol Woo; Kornbluth, Mordechai; Vandermause, Jonathan
- npj Computational Materials, Vol. 7, Issue 1
Works referencing / citing this record:
Artificial Intelligence based Autonomous Molecular Design for Medical Therapeutic: A Perspective
preprint, January 2021
- Joshi, Rajendra P.; Kumar, Neeraj
- arXiv