Guiding Colloidal Refolding using Machine Learning

One of the most exciting frontiers of material science is to emulate nature’s ability to fold functional and reconfigurable structures – proteins, RNA – from basic elements. Recently, synthetic chains of micron-sized droplets were shown to successfully fold using DNA-mediated interactions and temperature protocols [1]. Controlling this folding is difficult: the number of folding pathways proliferates with the number of droplets. Here, we develop a reinforcement learning algorithm able to identify rare reliable pathways and to fold large chains into desired structures. Our algorithm uncovers strategies for folding a single chain into multiple structures, and switching between them, paving the way for systematic folding-based synthesis of functional materials.

[1] McMullen, A., Muñoz Basagoiti, M., Zeravcic, Z., Brujic, J.. Self-assembly of emulsion droplets through programmable folding. Nature 610, 502–506 (2022).