CHORISO - chemical reaction SMILES from academic journals

CHORISO (CHemical Organic ReactIon SMILES Omnibus) is a curated dataset containing chemical reactions SMILES extracted from high-impact factor journals. It is built using the CJHIF dataset, and the resulting data is used to propose a new holistic evaluation of reaction prediction models (see paper). A detailed explanation of the processing steps and proposed metrics is included in this repo.
The following files are included:
choriso_public.tar.gz: compressed file containing the ChORISO dataset, 2'224'239 canonical reaction SMILES.uspto_public.tar.gz: file containing the USPTO dataset cleaned and processed following the same pipeline than CHORISO.splits.tar.gz: compressed folder containing the training, validation and test files used to train and evaluate models in the study.