Group contribution-based property modeling for chemical product design: A perspective in the AI era

Abstract

We provide a perspective of the challenges and opportunities for the group contribution approach for property prediction modeling with respect to their use in the design of chemical-based products in the modern era of artificial intelligence. In particular, we discuss issues related to the correct formulation of the product design problem, representation of molecular structures for property prediction as well as generation of product candidates, regression of property model parameters, and the integration of property related data and models with product design methods and tools using several conceptual examples. The need for developing appropriate hybrid AI models is described and recommendations for future work are presented.