Polymer solution phase behavior is crucial to many aspects of polymer synthesis and design; however, many theoretical predictions disagree with experiments. Improvements to theoretical models such as Flory-Huggins requires extensive experimental data and yields chemistry-dependent results. Alternatively, data-centric machine learning models are increasingly demonstrating value for polymer property prediction and design. Herein, we discuss the development of a machine learning (ML) framework for the prediction of temperature-dependent phase behavior of polymer solutions using a database of cloud point temperatures. The structure and chemistry of the constituents, and the breadth of parameter space, are encoded via chain length, molecular descriptors, fingerprints, and/or Hansen solubility parameters. A single ML model is shown to predict linear polymer solution cloud point temperatures within 1-3 °C and capture various polymer-solvent phase behaviors such as upper critical, lower critical, and closed loop curves. Extrapolation to new polymer-solvent systems, as well as more complex macromolecular architectures (e.g.  stars and polymer grafted nanoparticles), requires as little as 20 cloud point temperatures, which can be further reduced with active learning methods. This confirms the common underlying physics, but the details are obscured in the ML model.  This hinders human interpretability and trust. Therefore, a hybrid framework is being developed to combine Flory-Huggins theory with these ML-models to (i) improve predictability and generalization with less experimental data, and (ii) improve interpretability via estimating theoretical parameters with ML.

Machine learning framework for estimating temperature-dependent phase behavior of polymer solutions

Where thousands of chemistry professionals meet to share ideas and advance scientific and technical knowledge. Spring 2023 abstracts, posters, and presentations will launch at the end of March.

ACS Spring 2023

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.