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 understand ing of chemical reactions through inductive learning and analogical re asoning by considering individual reactions. The advent of reaction da tabases supports the development of machine-learning techniques for th e automatic acquisition of knowledge on chemical reactions. Hierarchic al classification and generalization can collect similar reactions int o a reaction type and detect those features-functional groups and elec tronic 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 r eaction types as Michael additions, Friedel-Crafts alkylation, and fre e radical additions. The automatic classification of these reactions w as able to find these and other important reaction types, indicated th e scope and limitations of these reaction types, and discovered some m ore unusual processes.