Quantifying the effect of realistic variations in sequence-define macromolecule aggregate morphology with supervised machine learning

Block copolymers have the capacity to self-assemble into an assortment of intricate architectures, with applications in things like drug delivery and personalized medicine. New techniques in chemistry allow for increasing control over these architectures through sequence control at the monomer level rather than the block level, greatly expanding the horizons for the design of their properties. However, chemical synthesis is imperfect, and inherent stochasticity in synthesis and self-assembly poses a significant challenge for the precise modeling and control of these systems. While it is known that different monomer sequences result in different morphologies, predicting the exact response to variability with the monomer sequence is a challenging problem. In this work, we utilize an unsupervised machine learning method to quantitatively characterize the morphologies of self-assembled model copolymers. We find that the morphology response to sequence variability can be accurately predicted using neural networks, and we identify preferred pathways through the low-dimensional morphology space. This analysis sheds light on how sequence variability can be tolerated and perhaps even controlled to design sequence defined copolymers for technological applications.