Research on chemical reaction prediction model based on Fairseq
Tingting Wang · 2021
Predicting the products of chemical reactions is a conspicuous difficulty in organic chemistry. The model in this paper is used to predict the compounds produced under known reactants and reagents. The author uses fairseq framework of python to construct a neural machine translation model based on SMILES strings of chemical reactants, reagents and products. The molecular formula of the products of the reactions can be well predicted. After training, the loss and the ppl of the model are 0.435 and 1.35. The final two bleu ontest part are 12.71 and 12.60 respectively. Fairseq performs well in molecular prediction, which shows that the cross domain traditional neural machine translation method has broad prospects in chemistry, and the use of artificial intelligence method to assist traditional research has strong practicality.