MACHINE LEARNING The chemistry of errors

The application of machine learning to big data, to make quantitative predictions about reaction outcomes, has been fraught with failure. This is because so many chemical-reaction data are not fit for purpose, but predictions would be less error-prone if synthetic chemists changed their reaction design and reporting practices.

Automated summary

The application of machine learning to predict reaction outcomes using big data has encountered significant challenges. Many chemical reaction data are not suitable for accurate predictions, making it necessary for synthetic chemists to change their reaction design and reporting practices.