Assessment and validation of machine learning methods for predicting molecular atomization energies

Katja Hansen, Grégoire Montavon, Franziska Biegler, Siamac Fazli, Matthias Rupp, Matthias Scheffler, O. Anatole Von Lilienfeld, Alexandre Tkatchenko, Klaus Muller

Research output: Contribution to journalArticle

230 Citations (Scopus)

Abstract

The accurate and reliable prediction of properties of molecules typically requires computationally intensive quantum-chemical calculations. Recently, machine learning techniques applied to ab initio calculations have been proposed as an efficient approach for describing the energies of molecules in their given ground-state structure throughout chemical compound space (Rupp et al. Phys. Rev. Lett. 2012, 108, 058301). In this paper we outline a number of established machine learning techniques and investigate the influence of the molecular representation on the methods performance. The best methods achieve prediction errors of 3 kcal/mol for the atomization energies of a wide variety of molecules. Rationales for this performance improvement are given together with pitfalls and challenges when applying machine learning approaches to the prediction of quantum-mechanical observables.

Original languageEnglish
Pages (from-to)3404-3419
Number of pages16
JournalJournal of Chemical Theory and Computation
Volume9
Issue number8
DOIs
Publication statusPublished - 2013 Aug 13

Fingerprint

machine learning
atomizing
Atomization
Learning systems
Molecules
predictions
molecules
chemical compounds
Chemical compounds
Ground state
energy
ground state

ASJC Scopus subject areas

  • Physical and Theoretical Chemistry
  • Computer Science Applications

Cite this

Assessment and validation of machine learning methods for predicting molecular atomization energies. / Hansen, Katja; Montavon, Grégoire; Biegler, Franziska; Fazli, Siamac; Rupp, Matthias; Scheffler, Matthias; Von Lilienfeld, O. Anatole; Tkatchenko, Alexandre; Muller, Klaus.

In: Journal of Chemical Theory and Computation, Vol. 9, No. 8, 13.08.2013, p. 3404-3419.

Research output: Contribution to journalArticle

Hansen, K, Montavon, G, Biegler, F, Fazli, S, Rupp, M, Scheffler, M, Von Lilienfeld, OA, Tkatchenko, A & Muller, K 2013, 'Assessment and validation of machine learning methods for predicting molecular atomization energies', Journal of Chemical Theory and Computation, vol. 9, no. 8, pp. 3404-3419. https://doi.org/10.1021/ct400195d
Hansen, Katja ; Montavon, Grégoire ; Biegler, Franziska ; Fazli, Siamac ; Rupp, Matthias ; Scheffler, Matthias ; Von Lilienfeld, O. Anatole ; Tkatchenko, Alexandre ; Muller, Klaus. / Assessment and validation of machine learning methods for predicting molecular atomization energies. In: Journal of Chemical Theory and Computation. 2013 ; Vol. 9, No. 8. pp. 3404-3419.
@article{23a554c03a8946ce8448abc2da0f6977,
title = "Assessment and validation of machine learning methods for predicting molecular atomization energies",
abstract = "The accurate and reliable prediction of properties of molecules typically requires computationally intensive quantum-chemical calculations. Recently, machine learning techniques applied to ab initio calculations have been proposed as an efficient approach for describing the energies of molecules in their given ground-state structure throughout chemical compound space (Rupp et al. Phys. Rev. Lett. 2012, 108, 058301). In this paper we outline a number of established machine learning techniques and investigate the influence of the molecular representation on the methods performance. The best methods achieve prediction errors of 3 kcal/mol for the atomization energies of a wide variety of molecules. Rationales for this performance improvement are given together with pitfalls and challenges when applying machine learning approaches to the prediction of quantum-mechanical observables.",
author = "Katja Hansen and Gr{\'e}goire Montavon and Franziska Biegler and Siamac Fazli and Matthias Rupp and Matthias Scheffler and {Von Lilienfeld}, {O. Anatole} and Alexandre Tkatchenko and Klaus Muller",
year = "2013",
month = "8",
day = "13",
doi = "10.1021/ct400195d",
language = "English",
volume = "9",
pages = "3404--3419",
journal = "Journal of Chemical Theory and Computation",
issn = "1549-9618",
publisher = "American Chemical Society",
number = "8",

}

TY - JOUR

T1 - Assessment and validation of machine learning methods for predicting molecular atomization energies

AU - Hansen, Katja

AU - Montavon, Grégoire

AU - Biegler, Franziska

AU - Fazli, Siamac

AU - Rupp, Matthias

AU - Scheffler, Matthias

AU - Von Lilienfeld, O. Anatole

AU - Tkatchenko, Alexandre

AU - Muller, Klaus

PY - 2013/8/13

Y1 - 2013/8/13

N2 - The accurate and reliable prediction of properties of molecules typically requires computationally intensive quantum-chemical calculations. Recently, machine learning techniques applied to ab initio calculations have been proposed as an efficient approach for describing the energies of molecules in their given ground-state structure throughout chemical compound space (Rupp et al. Phys. Rev. Lett. 2012, 108, 058301). In this paper we outline a number of established machine learning techniques and investigate the influence of the molecular representation on the methods performance. The best methods achieve prediction errors of 3 kcal/mol for the atomization energies of a wide variety of molecules. Rationales for this performance improvement are given together with pitfalls and challenges when applying machine learning approaches to the prediction of quantum-mechanical observables.

AB - The accurate and reliable prediction of properties of molecules typically requires computationally intensive quantum-chemical calculations. Recently, machine learning techniques applied to ab initio calculations have been proposed as an efficient approach for describing the energies of molecules in their given ground-state structure throughout chemical compound space (Rupp et al. Phys. Rev. Lett. 2012, 108, 058301). In this paper we outline a number of established machine learning techniques and investigate the influence of the molecular representation on the methods performance. The best methods achieve prediction errors of 3 kcal/mol for the atomization energies of a wide variety of molecules. Rationales for this performance improvement are given together with pitfalls and challenges when applying machine learning approaches to the prediction of quantum-mechanical observables.

UR - http://www.scopus.com/inward/record.url?scp=84882415695&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84882415695&partnerID=8YFLogxK

U2 - 10.1021/ct400195d

DO - 10.1021/ct400195d

M3 - Article

AN - SCOPUS:84882415695

VL - 9

SP - 3404

EP - 3419

JO - Journal of Chemical Theory and Computation

JF - Journal of Chemical Theory and Computation

SN - 1549-9618

IS - 8

ER -