Virtual reaction condition optimization based on machine learning for a small number of experiments in high-dimensional continuous and discrete variables

We examined a virtual simulation scheme for reaction condition optimization using machine learning for a small number of experiments with nine reaction conditions, consisting of five continuous and four discrete variables. Simulations were performed for predicting product yields in a synthetic reaction of tetrasilabicyclo[1.1.0]but-1(3)-ene (SiBBE). The performances in terms of accuracy and efficiency in the simulations and the chemical implications of the results were discussed.