Predicting chemical reactions and mechanisms with computer expert systems

Underlying medicine and biology is the biochemistry of the cell, which in turn is based on the reactivity of organic molecules. Our understanding and application of organic chemistry, the very chemistry of life, has profound implications for many important problems and human diseases. This includes the design and discovery of pharmaceutical drugs, the vast majority of which are small organic molecules.;The ultimate goal of this work is to develop a computer system to solve fundamental problems in chemistry, with an interim goal of solving problems in synthetic organic chemistry at a level comparable to a human expert. The primary problem addressed here is predicting the results of arbitrary chemical reactions, a key issue in tasks ranging from synthesis design to reaction discovery.;Our first version of a reaction prediction expert system encodes expert knowledge in the form of a knowledge-base of over 1,500 manually composed rules. This knowledge-base enables robust predictions of detailed reaction mechanisms with a level of precision largely lacking in prior chemical information systems. Ultimately, this grants the system predictive power comparable to that of a chemistry graduate student.;Applications of this expert system towards retro-synthetic analysis and combinatorial library design are illustrated in the expansion of ChemDB, a large database of chemical information, to the scope of readily synthesizable virtual chemicals. The Reaction Explorer system further demonstrates the reaction prediction functions in the context of a computer-based learning system, which realizes interactive learning features never before possible for organic chemistry, including support for inquiry-based learning.;Despite the success of this rule-based expert system, the scope of the system is necessarily limited by the knowledge-base. To overcome these limitations and achieve greater potential predictive power, we have laid the groundwork for a new version of the system that is instead driven by general principles of physical organic chemistry. In different capacities, both versions of the expert system mimic the qualitative problem-solving ability of human experts, but with the speed, precision, and combinatorial power of an automated system.