Predicting Phase-Specific Entropy of Polymerization Using Machine Learning to Accelerate the Design of Recyclable Plastics

Today’s plastics are difficult to recycle, and therefore plastic waste accumulates in landfills, oceans, and other natural habitats. It is important to design novel intrinsically circular polymers (ICPs) that can be generated from renewable feedstocks with the hope of eliminating the end-of-life problem and pollution of today’s plastics. These novel ICPs can undergo a process called chemical recycling through which the quality of the starting polymer is maintained as it is broken down into its building blocks for reuse, allowing the material to theoretically be recycled infinitely. In order to expedite the design of such materials, we are developing a machine learning model that can predict a thermodynamic indicator of recyclability for a polymer-reaction environment pair. This machine learning model is trained on a dataset assembled for the first time of phase-specific entropies of polymerization with features engineered to represent the complexities in entropy of polymerization, a property that is notoriously difficult to capture accurately using molecular modeling methods. This model will be used to rapidly screen the vast molecular search space created by hybrid pathways for monomer candidates that have thermodynamic properties suitable for sustainable chemical recycling.