Using a MixSolDB, a curated solubility dataset for multicomponent solvent systems, a GNN-based framework with semi-supervised distillation from COSMO-RS data improved prediction accuracy and broadened applicability.
In this work, we report the development of multiple new machine learning (ML) models to accurately predict homopolymer/copolymer solubility over a diverse set of polymers & solvents, using explainable AI to provide polymer design recommendations.
Recyclable polymers that can be produced at scale and readily tuned within the same polymer framework for specific properties are critical to achieving a circular materials economy. To this end, synthetic poly(3-hydroxyalkanoate)s (PHAs) have emerged as high-performance, chemically recyclable variants of biological PHAs, but their difficult monomer syntheses and suboptimal recycling efficiencies pose challenges for large-scale deployment. In this study, we investigated a β-isopropylidene PHA, i -PHA, for which the lactone monomer can be synthesized by existing industrial methods from biomass-derived isobutyric acid. The alkylidene substituent prevents decarboxylative degradation typically observed during PHA depolymerization, enabling near-quantitative chemical recycling to monomer. Controlled hydrogenation of the β-isopropylidene side group produces PHAs with diverse performance metrics that are competitive with a range of commodity polymers, spanning strong fibers to ductile thermoplastics to superglue epoxy resins.