In previous work, we introduced a machine learning approach to predict productive mechanistic reactions.  The representations and methods presented allowed for practical mechanistic reaction prediction over the set of polar organic reaction.  Here we describe several improvements to our previous approach.  First, we describe extensions to the orbital interaction representation to cover pericyclic, radical, and stereospecific reactions.  Next we describe a new larger and more diverse training set of chemical reactions, derived from extending the use of the Reaction Explorer expert system along with manual curation from literature and graduate level texts.  Then, we show how our general machine learning approach, with the new representations and expanded dataset, exhibits excellent performance results.  Finally, a multi-step pathway prediction application is made available.

Learning to predict more chemical reactions: Model extensions and an expanded training set

243rd National Meeting (2012)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.