Neural Network Finds an Enzyme that Can Break Down Polyurethane

Why It Matters

By turning a major, cross‑linked plastic waste stream into a recyclable feedstock, the breakthrough could dramatically reduce landfill and incineration of polyurethane, opening a scalable, circular pathway for a $22 million‑ton market. It also validates neural‑network‑based enzyme engineering as a rapid route to solve other persistent polymer‑waste challenges.

Summary

Researchers used AI‑driven protein design tools, including the neural networks Pythia‑Pocket, Pythia and a new graph‑based model called GRASE, to engineer a novel enzyme that degrades polyurethane. The engineered enzyme displayed up to 30‑fold higher activity than the best natural counterpart, and when combined with diethylene glycol at 50 °C achieved over 450‑fold activity, breaking down 98% of polyurethane in 12 hours and 95%+ in kilogram‑scale trials. The process yields the polymer’s original monomers, enabling their reuse in fresh polyurethane production. The work, published in Science, demonstrates the power of AI‑guided enzyme design for tackling hard‑to‑recycle plastics.