Our co-founders' research article "ORDerly: Data Sets and Benchmarks for Chemical Reaction Data" has been published online in the Journal of Chemical Information and Modelling (JCIM). ORDerly is an open source pipeline for cleaning chemical reaction data sets stored in Open Reaction Database (ORD), and three new benchmarks were created: ORDerly-forward (forward prediction), ORDerly-retro (retrosynthesis), and ORDerly-condition (the first condition prediction benchmark).
Within ReactWise we are using ORDerly to create datasets that we use for pre-training our algorithms to enable efficient transfer-learning optimization and understand chemical reactions before performing the first experiment.
Read the full article on JCIM.