Accelerating Discovery of Liquid Crystal Polymeric Materials with Extreme Properties using Machine Learning

Liquid crystal polymeric materials are polymers combining the ordering properties of liquid crystals and rubbery elasticity from the polymer network. Liquid crystal polymers have demonstrated great potential as implants and medical devices due to the versatile stimuli responsiveness and programmable actuation. Here, we are interested in identifying new liquid crystal polymeric materials with extreme properties, such as body-temperature phase transition temperature, high fatigue resistance, and high actuation strain. The traditional way to identify materials with desired properties is intuition-driven, based on trial-and-test strategies, which is not only time-consuming but highly likely to miss the optimal candidates, especially under the scenarios that multiple “conflicting” properties need to meet at the same time. To address this issue, we inverse design of polymer with desired properties using machine learning methods. In particular, we use generative models to explore potential chemical space and use transfer learning to find the quantitative structure-property relationship (QSPR) based on a limited dataset. The methods developed here are highly generalizable to solve other polymer design problems from the molecular scale.