AUTOMATIC EXTRACTION OF CHEMICAL KNOWLEDGE FROM ORGANIC-REACTION DATA - ADDITION OF CARBON-HYDROGEN BONDS TO CARBON-CARBON DOUBLE-BONDS

Historically, chemists have gained a large portion of their understanding of chemical reactions through inductive learning and analogical reasoning by considering individual reactions. The advent of reaction databases supports the development of machine-learning techniques for the automatic acquisition of knowledge on chemical reactions. Hierarchical classification and generalization can collect similar reactions into a reaction type and detect those features-functional groups and electronic effects-that are necessary for a reaction type as well as those that are dispensable. HORACE, a system based on these principles, was applied to a dataset of 120 reactions involving the addition of a C-H bond to a C=C double bond, a reaction that comprises such important reaction types as Michael additions, Friedel-Crafts alkylation, and free radical additions. The automatic classification of these reactions was able to find these and other important reaction types, indicated the scope and limitations of these reaction types, and discovered some more unusual processes.