Unveiling Microgel Structure through Machine Learning

Microgel particles are a versatile type of soft particles consisting of crosslinked polymer chains. In particular, thermoresponsive microgels undergo a volume phase transition as a function of temperature changes, making them attractive for various technological applications such as drug delivery and art restoration. Their structure and elastic properties closely depend on the amount of crosslinker monomers used in their synthesis and their distribution throughout the particle. However, experimental characterization of crosslinker structural features is a cumbersome task.

Here, I will show that by using radial density profiles from monomer-resolved molecular simulations of microgel particles under different crosslinker conditions as input data for machine learning models, in particular autoencoders and neural networks, we can predict the crosslinker concentration and estimate its spatial distribution, respectively. Finally, with this structural insight, we provide a prediction of the shrinkage behavior, with good agreement with experimental data.

Our results provide a powerful method for predicting structural behavior from minimal information, which can be easily extended to experimental data. This opens the door to the inverse design of tailored structures.