Machine Learning for Polymer Swelling in Liquids

Swelling in liquids is of paramount importance for polymers used in many liquid-phase applications. This critical property has motivated numerous analytical theories and empirical experiments as well as recent atomistic simulations; however, a data-driven approach for polymer swelling is currently not available. In this study, we develop a machine learning (ML) methodology to investigate polymer swelling in liquids. This methodology is illustrated for the swelling of organic solvent nanofiltration (OSN) membranes and polydimethylsiloxane (PDMS) in various solvents. First, chemically intuitive descriptors such as solubility parameters and solvent properties are proposed to construct ML models. Using kernel ridge regression, the model based on the solubility parameters of the solvent and polymer is found to offer the best quantitative prediction and reveal multimodal swelling behavior for OSN membranes. For PDMS swelling, the solubility parameter and geometry of solvent are identified to be key properties. Then, a molecular representation via the sum-of-fragments approach is proposed and demonstrated remarkable predictive capability. Through appropriate data augmentation, reasonable out-of-sample prediction is achieved for polyetherimide swelling in nine solvents and PDMS swelling in substituted aromatic solvents. Finally, principal component analysis is applied to the proposed sum-of-fragments to explore its suitability as a molecular representation and the chemical space of polymer swelling. The relationships between molecular fragments and swelling degrees are quantitatively determined by Pearson correlations. This ML study demonstrates the development and utilization of physically meaningful chemical descriptors to construct models capable of superior prediction and unraveling fundamental insight into polymer swelling. Such a methodology can also be extended to other physical properties for polymers in liquids, thereby expanding its scope of potential applications.