Physical Sciences Division
Published July 20, 2021 | Version v1
Journal article Open

A community-powered search of machine learning strategy space to find NMR property prediction models

Creators

  • 1. University of Bristol
  • 2. University of Chicago
  • 3. Carnegie Mellon University
  • 4. Korea Institute of Science and Technology Information
  • 5. BNP Paribas Cardif
  • 6. Fyusion, Inc.
  • 7. Kaggle
  • 8. eBay
  • 9. Bosch Research and Technology Center
  • 10. Korea Advanced Institute of Science and Technology
  • 11. MINDs n Company
  • 12. Totient
  • 13. KAIST
  • 14. University of Chicgo
  • 15. Medbravo.org
  • 16. Bosch Center for Artificial Intelligence

Description

The rise of machine learning (ML) has created an explosion in the potential strategies for using data to make scientific predictions. For physical scientists wishing to apply ML strategies to a particular domain, it can be difficult to assess in advance what strategy to adopt within a vast space of possibilities. Here we outline the results of an online community-powered effort to swarm search the space of ML strategies and develop algorithms for predicting atomic-pairwise nuclear magnetic resonance (NMR) properties in molecules. Using an open-source dataset, we worked with Kaggle to design and host a 3-month competition which received 47,800 ML model predictions from 2,700 teams in 84 countries. Within 3 weeks, the Kaggle community produced models with comparable accuracy to our best previously published 'in-house' efforts. A meta-ensemble model constructed as a linear combination of the top predictions has a prediction accuracy which exceeds that of any individual model, 7-19x better than our previous state-of-the-art. The results highlight the potential of transformer architectures for predicting quantum mechanical (QM) molecular properties.

Data availability

All data is available from the paper, its Supporting Information files, and the following repositories: http://osf.io/kcaht http://github.com/larsbratholm/champs_kaggle.

Files

journal.pone.0253612.pdf

Files (4.2 MB)

Name Size Download all
Article
md5:9b95b864ca3a233b73cfca9d0acd4226
1.9 MB Preview Download
Appendix
md5:e167bceed3b3eb2eba2db87de8fcdd9a
2.3 MB Preview Download

Additional details

Identifiers

DOI
10.1371/journal.pone.0253612
Other
oai:uchicago.tind.io:5951

Funding

EPSRC
National Productivity Investment Fund (NPIF) for Doctoral Studentship funding
EPSRC
EP/N510129/1
Leverhulme Trust
Philip Leverhulme Prize
Royal Society
URF/R/180033
EPSRC
"CHAMPS" programme
National Research Foundation of Korea
2018R1D1A1B07049981
National Research Foundation of Korea
2019M3E5D4065968
Ministry of Science
ICT

UChicago Information

Division(s)
Physical Sciences Division
Department(s)
Computer Science