(13e) Machine-Learning on the Coarse-Grained Polymer Genome
The recent emergence of machine-learning (ML) techniques in tandem with molecular modeling has greatly enhanced simulation capabilities. The possibility of using existing and/or generated data within ML frameworks to greatly accelerate property prediction is particularly intriguing for materials design. While there has been significant advancement and progress in the context of small molecules or crystal structures, this objective is notably difficult for polymers or soft materials, for which target properties are often characterized by distributions resulting from an ensemble of configurations governed by subtle interactions. In this talk, I will discuss the viability of using coarse-grained simulations in tandem with supervised machine-learning to predict polymer properties based on polymer sequence, i.e., to define the chemistry/structure->property relationships that characterize the polymer genome. Within this framework, we systematically interrogate the influence polymer composition and topology on observed polymer properties for a defined coarse-grained chemical space. I will further show how this framework lends itself to generative design strategies to achieve target properties. Overall, this work aims to highlight the promising capabilities (and difficulties) in using ML towards soft materials design.