An expert system for organic synthesis using automated reasoning (theorem prover, retrosynthesis, logic, representation, and/or graph, knowledge)
Tunghwa Wang · 1986
The problem of planning organic syntheses is a problem well-known to require deep reasoning because of its complexity and the large and constantly increasing knowledge base. A new approach based on a logic representation of knowledge is used to attack this problem, because this kind of formulation of knowledge is both powerful and flexible. The inference engine of this system is a general purpose theorem prover built upon LMA (Logic Machine Architecture), a package of Pascal routines developed at Argonne National Laboratory. In this way, the problem of organic synthesis is essentially reduced to a problem of a controlled sequence of automated reasoning processes. A backward chaining control process is applied to build a problem-solving graph in which the nodes are molecules and each edge is an organic reaction. This problem-solving graph is generated in two phases. In the first phase organic reaction paths of multiple-step pseudo reactions are used to generate intermediate molecules of interest instead of real single-step reactions. In the second phase the multiple-step sequences proposed in the first phase are verified. This two phase scheme reduces the search space and allows parallel processing of tasks from both phases.