Informatics Pipeline to Identify Polymers Capable of Controllable Depolymerization

Enthalpy of polymerization is a key thermodynamic property to determine if a polymer can be controllably degraded to monomer feedstock at the end-of-use, a capability essential to mitigating the global polymeric waste problem. Due to the favorable thermodynamics of ring opening polymerizations (ROP) in creating depolymerizable polymers, the enthalpy of ROP (ΔH^ROP) is the focus of this study. Thanks to a recently developed first-principles computational scheme, a high-quality database for computed ΔH^ROP has been developed in order to help overcome the issue of data scarcity present in the current literature. Using this computational database, along with the limited experimental data present in literature, a data-fusion approach has been adopted to create a machine learning (ML) model capable of predicting experimental ΔH^ROP near chemical accuracy, ~5 kJ/mol. This ML model is then used in conjunction with the first-principles computational method to create a pipeline where the ML method screens the massive chemical space that makes up ROP chemistries, and the most promising candidates are then passed to the first-principles calculations for validation. This iterative process exploits the ML model uncertainty as well as discrepancies between ML predictions and first-principles calculations to create active learning cycles used to optimize the generalizability of the ML model. This informatics pipeline provides an important step towards creating an automated process for the discovery and design of depolymerizable polymers, meant specifically to create plastics that can truly be sustainable.