Prediction of chemical reaction yields using deep learning

Abstract Artificial intelligence is driving one of the most important revolutions in organic chemistry. Multiple platforms, including tools for reaction prediction and synthesis planning based on machine learning, have successfully become part chemists’ daily laboratory, assisting domain-specific synthetic problems. Unlike retrosynthetic models, yields has received less attention spite enormous potential accurately predicting conversion rates. Reaction describing percentage reactants converted to desired products, could guide chemists help them select high-yielding reactions score routes, reducing number attempts. So far, yield predictions been predominantly performed high-throughput experiments using a categorical (one-hot) encoding reactants, concatenated molecular fingerprints, or computed chemical descriptors. Here, we extend application natural language processing architectures predict properties given text-based representation reaction, an encoder transformer model combined with regression layer. We demonstrate outstanding performance two experiment sets. An analysis reported open-source USPTO data set shows that their distribution differs depending mass scale, limiting applicability predictions.