Designing an Expert System for Organic Synthesis
Peter Y. Johnson, Ilene Burnstein, John F. Crary, Martha Evans, Tunghwa Wang · ACS symposium series · 1989
SYNLMA, an expert system for organic synthesis, with a theorem prover as its inference engine and NCI's XTCHEM as its user interface, uses a retrosynthetic approach to find reaction pathways and generate a problem-solving tree representing the alternative designs it has explored. Presently, the system is capable of handling compounds of the order of complexity of Darvon, Ibuprofen, and the bicyclic system, cocaine. The combinatorial explosion that results from the input of larger target molecules has convinced us of the need for strategic planning during the synthesis process. We have developed a three-stage approach to aid SYNLMA in the planning process. The first stage identifies abstracted potential starting materials or name reaction derived synthons using graph overlay techniques to compare them with complex substructures in the target molecule. The second stage involves using "PMCD" strategies to define graphical paths between the target and abstracted synthons or starting materials. The leaf nodes on this path represent "chemical islands" which are then connected by general reaction rules. The third stage defines the tree by supplying specific reaction rules.