Google Scientists Use AI to Design a Plastic-Digesting Enzyme

Scientists have successfully used artificial intelligence to design a new enzyme capable of breaking down plastic polymers. This AI-designed enzyme, developed using RFDiffusion and PLACER, represents a significant step toward engineering proteins that can catalyze reactions beyond what is found in nature.

The Challenge of Designing Enzymes for Plastic Breakdown

Enzymes are powerful biological catalysts, enabling a wide range of chemical reactions with precision. However, many reactions—such as digesting plastics or converting carbon dioxide into useful compounds—lack natural enzymatic solutions. While directed evolution has helped modify existing enzymes, creating entirely new ones remains a challenge.

With AI-driven protein design, researchers now have the ability to develop enzymes from scratch. In this study, scientists created an enzyme capable of breaking ester bonds, a chemical structure found in many biological molecules and synthetic materials like polyester plastics (PET).

How AI Designed a Multi-Step Enzyme

Ester bonds link molecules together and are found in both natural biopolymers and synthetic plastics like PET (polyethylene terephthalate). Breaking these bonds is essential for plastic degradation and recycling, as well as for various industrial and biochemical processes. Specifically, enzymes capable of breaking ester bonds could:

• Aid plastic degradation – Helping recycle materials like PET.

• Advance industrial chemistry – Creating sustainable alternatives for chemical production.

While many enzymes exist that can break ester bonds in biological systems, designing one that can efficiently break down synthetic plastics is a complex challenge. The reaction involves multiple steps:

Binding to the ester bond

• Breaking the bond efficiently

• Releasing the broken-down molecules without getting stuck

To tackle this challenge, researchers used AI-driven protein design, relying on two key systems:

• RFDiffusion – An AI tool that generates protein structures by analyzing the average positions of amino acids found in natural ester-breaking enzymes. It uses a randomized approach to explore diverse protein designs, ensuring that the generated enzymes have the potential to bind to and interact with ester bonds effectively. The results were then fed to another neural network, which selected amino acids to form a pocket capable of holding an ester that breaks down into a fluorescent molecule—allowing researchers to track enzyme activity based on its glow.

• PLACER – A generative AI trained on structural protein data, designed to refine enzyme configurations. After initial enzyme structures were generated, PLACER helped adjust amino acid positioning to ensure enzymes could cycle through multiple reaction steps rather than stalling after a single use. The AI was trained by introducing random distortions to known protein structures, forcing it to learn how to restore functional configurations. This process enabled PLACER to capture key structural details that allow enzymes to adopt different conformations throughout the reaction sequence, mimicking how natural enzymes operate.

Refining the AI-Generated Enzymes

Initial designs produced 129 enzyme candidates, but only two showed fluorescence, indicating catalytic activity. To improve results, the team incorporated PLACER, which refined enzyme structures and helped them cycle through multiple reaction steps instead of stalling.

The refined approach tripled the number of active enzymes, but many still stalled after a single reaction. Further AI screening with PLACER identified structures that could adopt key intermediate states, leading to:

• 18% of designed enzymes successfully cleaving ester bonds

• Two enzymes ("super" and "win") completing multiple reaction cycles

Creating an Enzyme That Can Digest Plastic

Through iterative AI-driven optimization, researchers eventually designed an esterase capable of breaking down PET plastic bonds. The final enzyme matched the efficiency of natural enzymes, demonstrating AI’s potential to engineer new biocatalysts for environmental applications.

What This Means

AI is transforming protein engineering, enabling the design of custom enzymes with real-world applications. One major impact is plastic degradation—engineered enzymes could improve recycling and help reduce plastic pollution by efficiently breaking down PET and other synthetic polymers.

Beyond environmental benefits, AI-designed enzymes could revolutionize industrial chemistry by replacing harsh chemical catalysts with sustainable, enzyme-driven alternatives. This could lead to greener processes in biofuel production, pollution cleanup, and drug development.

Although AI-assisted enzyme design remains complex, the ability to perform much of the work computationally—rather than relying on trial-and-error in the lab—marks a significant advancement. As AI tools improve, scientists may unlock entirely new biochemical processes with applications in medicine, agriculture, and sustainable manufacturing—reshaping industries and addressing global challenges.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.