Discovery of Chemical Transformations with the Use of Machine Learning

Abstract

Abstract. Research activities in the computer reaction modeling and synthesis design have led to the development of a number of computer systems that are successively used as a research tools to study the organic synthesis problems. One of such systems is the CSB (Chemical Sense Builder), which combines four different logic-and knowledge-based models to generate possible structural modes of chemical conversions. In these paper we briefly discuss the architecture and specially the learning methods implemented in the CSB. They exploit different sources of knowledge and chemical reaction data for the classification, generalization, and derivation of rules in a form of the graph transformation schemes. Any transformation is general description of a collection of similar reactions from the reaction data base. Application of these learned rules in the course of reaction simulation enables us to predict -for a given set of reacting molecules -all possible reaction courses, having responses in real chemistry.