Abstract
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.
Publication types
-
Research Support, Non-U.S. Gov't
MeSH terms
-
Algorithms
-
Citizen Science / methods*
-
Citizen Science / trends*
-
Community Participation
-
Forecasting / methods*
-
Humans
-
Machine Learning / trends
-
Magnetic Resonance Imaging / methods
-
Magnetic Resonance Spectroscopy / methods
-
Models, Statistical
Grants and funding
WG is partially supported by the EPSRC National Productivity Investment Fund (NPIF) for Doctoral Studentship funding. LAB thanks the Alan Turing Institute under the EPSRC grant EP/N510129/1. DRG is supported by the Leverhulme Trust (Philip Leverhulme Prize) and Royal Society (URF/R/180033). LAB and DRG acknowledge support of this work through the “CHAMPS” EPSRC programme grant EP/P021123/1. SC was supported by National Research Foundation of Korea (2018R1D1A1B07049981, 2019M3E5D4065968) funded by the Ministry of Science and ICT. Authors SB, LD, PH, AH, SK, ZK, MK, YL, JPM, TTN, MP, GR, WR, LS, NT, and DW are affiliated with commercial companies. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section.