Using Physical Property Models and Artificial Intelligence to Design Chemical Products

Designing chemical products is the process of combining pieces, e.g., molecular fragments or ingredients, into an assembly, e.g., chemical structures or mixture formulations, whose properties satisfy a set of design constraints. We explain how computational techniques, specifically artificial intelligence techniques, can greatly assist this design process. We demonstrate how proper representation of knowledge is essential for enabling the computer to manipulate substructures, how combinatorial algorithms are used to generate structures and enumerate isomers, how rule-based systems help select the estimation techniques needed to test each generated candidate chemical, and how machine learning can be used to improve the models used to find promising candidates.