Machine-learning-assisted Materials Discovery using Failed Experiments

Exploratory synthesis often entails educated guesswork and failed experiments. We demonstrate a machine learning approach to using data from failed experiments to target exploration and uncover relationships between physicochemical properties and reaction outcomes in crystallization of templated vanadium selenites. Our machine-learning based system predicts the outcome of candidate syntheses and recommends which reactants would produce the most "novel" outcomes (Raccuglia et al., Nature 2016).