Architecture of the models and functional validation of GRASE-identified enzymes. Credit: Science (2025). DOI: 10.1126/science.adw4487

As the use of AI spreads through every industry and becomes more of a part of our lives every day, researchers are also looking into ways it can be used to solve some of the world’s biggest problems. One of these problems is the world’s reliance on plastics for making everything from clothing to medical supplies to food wrappers, which is creating a massive amount of non-biodegradable waste—with more and more piling on every day. Much of this ends up wreaking havoc on various ecosystems and creating an overabundance of microplastics that end up in our food and water supplies.

Clearly, there is a need for recycling these materials. However, plastics remain one of the most difficult materials to recycle efficiently. But now, a team of researchers might have found a way to facilitate the process with the help of AI. Their study, published in Science, details how a neural network helped them find enzymes that can break down plastics faster and more efficiently than any they’ve found on their own.

The plastic recycling problem

Most plastic is never recycled, as the global recycling rate remains at only around 9%. Some plastics are easier to recycle than others. For example, polyurethane (PU) plastics—commonly used in flexible foams, like those found in running shoes, and in adhesives, coating and insulation—are a more challenging material to recycle. Yet, in 2024, 22 million metric tons of PU plastics were consumed globally.

Recycling thermoset PUs is challenging due to their cross-linked structure and stable urethane bonds. Glycolysis—a process in which a material is broken down into its constituent monomers using ethylene glycol—is currently the main industrial recycling method, but a large fraction of the leftover products are essentially just unusable, hazardous waste.

“Unlike thermoplastics such as polyethylene terephthalate (PET), which can be remolded into amorphous states through melt extrusion, thermoset polyurethanes cannot be reshaped to enhance susceptibility to enzymatic attack. Therefore, alternative recycling techniques are needed to overcome the depolymerization challenges posed by thermoset materials,” the study authors explain.

Searching for plastic-degrading enzymes

In order to fully recycle PU waste using glycolosis, certain enzymes are required. Specifically, robust, efficient enzymes that can operate in harsh, solvent-rich environments are needed to enable full recycling of PU waste, but these are hard to identify. Researchers must search through available literature, looking for enzymes that may have properties that can break down PU under specific conditions. This can take a lot of time. Then, the enzymes must be tested out.

This is exactly what the team did, and their initial results were less than satisfying. Out of 14 enzymes picked out of the literature, three worked well enough against the polymer they tested, and only one, Aes72, emerged as a promising candidate for further development in the end. Still, the activity and solvent compatibility of Aes72 were somewhat lacking.

AI to the rescue?

Then, the researchers turned to AI. Specifically, a graph neural network (GNN)-based framework for discovering active and stable enzymes, called GRASE, was used to identify more suitable enzymes for PU recycling, which were then prioritized on the basis of their sequence or structural identity compared to Aes72.

GRASE identified 24 top-ranked candidates, which were then validated in the lab using commercial PU foam and glycolysis-derived waste. While several of the enzymes performed well, one in particular stood out. The enzyme, referred to as AbPURase, had higher activity than previous enzymes and its structural analysis revealed features that confer stability and efficiency in harsh solvents.

When tested, AbPURase enabled 95% depolymerization of commercial PU foam at kilogram scale within eight hours and 98% at 12 hours. At a higher enzyme loading, 98.6% depolymerization was achieved after eight hours. The study authors call the performance of the neural network tool in picking out AbPURase “outstanding.”

The study authors also point out the surprising range of usefulness of some enzymes. They say, “Enzymes such as AbPURase, which are traditionally classified as esterases, were found to be primarily urethanases in this study.

“This observation aligns with the broader body of research on enzyme promiscuity and moonlighting, suggesting that many enzymes, rather than being strictly specific, have a range of activities that have often been obscured by conventional classification methods. Therefore, the full functional potential of enzymes may be significantly underappreciated in current annotations.”

Hope for the future of plastic recycling

This study brings forth a promising new potential for future plastic recycling endeavors to address the growing problem of plastic waste, especially from hard-to-recycle foams used in furniture, insulation, and cars. It also demonstrates the potential for AI in alleviating some of the environmental challenges we are currently faced with.

Still, the authors note that further protein engineering is needed to optimize enzyme performance and repolymerization of recycled monomers and continued use of tools, like GRASE, can unveil additional uses of previously underutilized enzymes.

Written for you by our author Krystal Kasal, edited by Lisa Lock, and fact-checked and reviewed by Robert Egan—this article is the result of careful human work. We rely on readers like you to keep independent science journalism alive. If this reporting matters to you, please consider a donation (especially monthly). You’ll get an ad-free account as a thank-you.

More information: Yanchun Chen et al, Glycolysis-compatible urethanases for polyurethane recycling, Science (2025). DOI: 10.1126/science.adw4487

© 2025 Science X Network

Citation: AI-guided enzyme discovery enables 98.6% breakdown of polyurethane foam in hours (2025, November 3) retrieved 3 November 2025 from https://phys.org/news/2025-11-ai-enzyme-discovery-enables-breakdown.html

This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.