Organic Synthesis as Artificial Intelligence Planning
Arman Masoumi, Mikhail Soutchanski, Andrea Marrella · 2013
Abstract. We explore advantages that can be gained from using expressive logic languages for semantic modelling of chemical reactions. First, we present a novel approach for logical representation of notions in organic chemistry, as well as for reasoning about generic chemical reactions. Subsequently, using this new semantic modeling of reactions, we explore what reasoning problems can be solved. We focus on solving organic chemistry synthesis problems, where the goal is to synthesize the target molecule from a set of starting stage molecules. We argue that this problem can be reduced to a planning problem in Artificial Intelligence. We conduct experimental study including empirical assessment of a PROLOG planner and two state-of-the-art planners. We investigate if they are capable of solving a set of instances of the organic synthesis problem. We report numerical data from our study and do comparative analysis of the planners. The novelty of our work is in using state-of-the art planners for solving the organic synthesis problem. The significance of our work is in methodology that we developed and in showing that expressive logical language can be useful for semantic modeling.