Exploratory synthesis often entails educated guesswork and innumerable failed experiments. We demonstrate a machine learning approach to using the data from failed experiments to intelligently target our exploration and uncover the relationship between physicochemical properties and reaction outcomes in the crystallization of templated vanadium selenites. This talk will review some common approaches to computationally aided materials science and exploratory synthesis, show where our work fits into this space, and discuss how we can use failure data and machine learning tools to be more effective scientists.
We digitized nearly a decade's worth of lab notebooks and explore how to use this previously unavailable data to improve our lab's efficiency. We built a machine-learning based system to predict the outcome of candidate syntheses and make recommendations about which reactants would produce the most "novel" outcomes. After building the system, we explored how to use it to gain greater insight into the mechanism of action, and built a tool that helped us identify novel hypotheses about the relationship between physicochemical properties and synthesis outcome. When carrying out hydrothermal synthesis experiments using previously untested, commercially available organic building blocks, our machine-learning model outperformed traditional human strategies, and successfully predicted conditions for new organically templated inorganic product formation with a success rate of 89 per cent .
 Raccuglia, P. et. al. Machine-learning-assisted materials discovery using failed experiments. Nature 533, 73–76 (2016)