An artificial neural network to predict reactivity ratios in radical copolymerization

Monomer reactivity ratios are central to our understanding of the polymerization rate, copolymer composition and sequence distribution of copolymers produced by radical polymerization but their measurement is not trivial. Although a number of different methods exist for the prediction of reactivity ratios of different monomer pairs, they have severely limited accuracy and are therefore rarely used in practice. In this work, we describe the use of an artificial neural network model that is capable of predicting reactivity ratios solely based on the chemical structures of the monomers. To train the model, a dataset of more than 5000 monomer pairs is used, with a molecular fingerprint of the monomers involved in the copolymerization as input to the model. It is demonstrated that the model has significantly higher accuracy than classical approaches such as the Q-e scheme and can be used to predict reactivity ratios for monomer pairs for which no kinetic data is available.