An Artificial Intelligence that Discovers Unpredictable Chemical Reactions

08 September 2020, Version 1
This content is a preprint and has not undergone peer review at the time of posting.


We present an artificial intelligence, built to autonomously explore chemical reactions in the laboratory using deep learning. The reactions are performed automatically, analysed online, and the data is processed using a convolutional neural network (CNN) trained on a small reaction dataset to assess the reactivity of reaction mixtures. The network can be used to predict the reactivity of an unknown dataset, meaning that the system is able to abstract the reactivity assignment regardless the identity of the starting materials. The system was set up with 15 inputs that were combined in 1018 reactions, the analysis of which lead to the discovery of a ‘multi-step, single-substrate’ cascade reaction and a new mode of reactivity for methylene isocyanides. p-Toluenesulfonylmethyl isocyanide (TosMIC) in presence of an activator reacts consuming six equivalents of itself to yield a trimeric product in high (unoptimized) yield (47%) with formation of five new C-C bonds involving sp-sp2 and sp-sp3 carbon centres. A cheminformatics analysis reveals that this transformation is both highly unpredictable and able to generate an increase in complexity like a one-pot multicomponent reaction.


Reactivity explorer


Comments are not moderated before they are posted, but they can be removed by the site moderators if they are found to be in contravention of our Commenting Policy [opens in a new tab] - please read this policy before you post. Comments should be used for scholarly discussion of the content in question. You can find more information about how to use the commenting feature here [opens in a new tab] .
This site is protected by reCAPTCHA and the Google Privacy Policy [opens in a new tab] and Terms of Service [opens in a new tab] apply.