SYNCHEM is a large knowledge-based domain-specific heuristic problem-solving system which has synthetic organic chemistry as its problem domain. It was designed to offer the practicing synthetic organic chemist a tool for discovering valid synthesis routes for organic molecules of interest without requiring that he provide online guidance, and also to serve as a vehicle for the application of artificial intelligence methodology in general, and heuristic programming and machine learning techniques in particular, to real-world problems in the natural sciences. The SYNCHEM project, initiated in the early seventies, was motivated by our conviction that problems selected to provide an environment for artificial intelligence research ought to be of consequence in themselves in order that the final results be worth the pain of pursuit. Some of the synthesis routes proposed by the SYNCHEM system have been original, and have been considered suggestive and useful by chemists who have experimented with the system.

As presently defined, SYNCHEM's task is to discover a provisional and plausible sequence of transformations (organic synthetic reactions) that will take the initial problem configuration (available starting materials) into the desired final state (the target molecule). SYNCHEM approaches this task by executing a retrosynthetic (backward-chaining) search in a problem state space characterized as a directed graph, where the nodes (the problem states) represent organic molecular structures and each arc specifies a known synthetic organic reaction that could, if relevant constraints and conditions are satisfied, transform the molecule at one incident node into that at the other. Even if the search has been successful, a number of additional events must occur before it may be legitimately claimed that the target compound has been synthesized. Laboratory validation of the discrete steps in the pathway is essential, for even if the system's KNOWLEDGE BASE fully reflects all that has been published concerning the chemistry of the selected reactions, what is known is almost never complete. No matter how plausible the individual instantiation of a well-established reaction generalization, it is rarely the case that the specific circumstances and environment of that particular application of the reaction transform will have been encountered and studied in the past. Since undesired reaction byproducts are often produced along with the target product, these must be identified and procedures must be devised to separate them out when necessary. In addition, the relative economics for alternative routes to the same goal must be estimated and weighed. Starting materials vary widely in cost and availability, as do the technologies and apparatus for carrying out synthesis procedures. The constraints on the search for a total synthesis of milligram quantities of a high purity bioactive substance needed for an investigation of its pharmacological properties are clearly not the same as those that obtain when the target is an industrial solvent for use by the tankcar load. While automatic wet bench laboratory validation is not a forseeable enhancement to the SYNCHEM system, it is not at all unreasonable to expect that a more fully developed version of the program will assume responsibility for the optimization of discovered pathways, close estimation of yields, prediction and specification of treatment procedures for reaction byproducts, and analysis of the resulting synthesis proposals for conformity with the requirements and constraints imposed on the problem solution.