Predicting self-assembly of sequence-controlled copolymers with stochastic sequence variation

Abstract

Sequence-controlled copolymers have the capacity to self-assemble into a wide assortment of complex architectures, with exciting applications in nanofabrication and personalized medicine. However, polymer synthesis is notoriously imprecise, and stochasticity in both chemical synthesis and self-assembly poses a significant challenge to tight control over these systems. While general design rules for blocky sequences are well understood, the influence of specific chemical sequences on self-assembled architecture is more challenging; variability within those sequences leads to an exceptionally complex (high-dimensional) design problem, which introduces an intractably high computational cost for brute force search. In this work, we utilized unsupervised learning to characterize the morphologies of nearly $15 \, 000$ sequence-controlled copolymer aggregates observed in Molecular Dynamics simulations. We found that sequence variation leads to relatively smooth trends in morphology in the same order parameters learned from monodisperse chains. Furthermore, structural response to sequence variation was accurately modeled using supervised learning, revealing several interesting trends in how specific families of sequences break down as monomer sequences become less homogeneous. Our work presents a way forward in understanding and controlling the effect of sequence variation in sequence-controlled copolymer systems, which can hopefully be used to design advanced copolymer systems for technological applications in the future.