Recent Advances on AI for Polymer Design

Advanced optimization and machine learning algorithms offer considerable promise for design of polymeric materials. The sequence space that is available for polymer design is extraordinarily large, and a central question that arises is whether it is possible to develop strategies that rely on a limited data set to train networks that can then be used to not only predict the properties of molecules having a given sequence, but to generate sequences that lead to desirable properties. In this work, that question is addressed by relying on simulations of model polymers that include different backbone and side groups to generate a numerical data base of molecules with different sequences. We also consider the extent to which such approaches can be extended to realistic polymeric materials, and the challenges that one faces in representing polymeric materials in a manner that is tractable and effective for polymer design.