Dataset for A Reactivity First Approach to Autonomous Discovery of New Chemistry

Dataset for an artificial intelligence which was able to learn how to identify chemical reactivity in the widest sense using deep learning. When this was used to control a robot, it was able to autonomously search many different reactions to assess the reactivity of reaction mixtures, including unknown datasets regardless the identity of the starting materials. The system not only found known-knowns (things we know) and known-unknows (new molecules predictable from known reactions), but also unknown-unknowns (new reactions that yielded new molecules). This was achieved within a budget of 15 inputs combined in 1018 reactions which allowed us to not only discover a new photochemical reaction, but also an unprecedented reactivity mode for a well-known reagent (p-Toluenesulfonylmethyl isocyanide, TosMIC). Remarkably, this involved the reaction of six equivalents of TosMIC in a ‘multi-step, single-substrate’ cascade reaction 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. Analysis reveals that this transformation is intrinsically unpredictable demonstrating that reactivity-first approaches could allow the discovery of unknown reaction methodologies.