A protein improved with machine learning helped researchers dismantle used plastic packaging and turn the recovered material into new PET. The enzyme, FAST-PETase, became a tangible demonstration of AI contributing to recycling chemistry when the University of Texas at Austin team published its work in 2022.[2]
PET—polyethylene terephthalate—is used in drinks bottles, food containers and polyester textiles. An enzyme can cut its long molecular chains into smaller components that can be recovered for reuse. That provides a route back to the ingredients of a material, with the possibility of building a new polymer from them.[4]
A few changes made the enzyme more useful
The starting point was a PET-degrading enzyme. The machine-learning contribution was to identify promising changes to its amino-acid building blocks. FAST-PETase combined three predicted substitutions with two already present in its parent scaffold, and the resulting enzyme showed improved activity over the tested alternatives between 30°C and 50°C.[1]
The underlying approach looks at a protein in three dimensions. A neural network learns which amino acids tend to fit particular local chemical surroundings. It can then flag positions where a different building block may fit better. Researchers make the proposed variants and measure whether their behaviour improves.[3]
That is a useful form of assistance because changing one small part of a protein can affect its stability or activity in ways that are hard to anticipate. The model narrows an enormous set of possible substitutions to candidates that deserve experimental attention. The final useful enzyme emerges from prediction combined with measurement.[3]
The original FAST-PETase demonstrations
| Material or step | Reported result | Conditions that define the result |
|---|---|---|
| PET from 51 thermoformed consumer products | Almost complete degradation within one week | Small punched-out samples; 50°C with enzyme replenishmentSource 1 |
| An entire commercial water bottle | Depolymerisation demonstrated | Thermal pretreatment; 50°C with fresh enzyme solution dailySource 1 |
| Recovered breakdown material | PET synthesised again | A laboratory recycling demonstrationSource 1 |
The distinction between the samples matters. A thin food container, a textile fibre and a bottle can share a polymer name while having different physical structures. The original experiments included several forms of PET and showed that the preparation of the material is part of the recycling process.[2]
Giving the enzyme a partner
Earlier research on PET-degrading enzymes explains a helpful division of labour. PETase attacks the polymer and releases smaller molecules. A partner called MHETase can act on one of those intermediate products. Bringing the enzymes together helps continue the breakdown towards reusable chemical ingredients, rather than allowing the process to stop partway.[4]
In a 2025 study, Siddhant Gulati and Qing Sun assembled FAST-PETase and MHETase on a scaffold displayed on the surface of yeast cells. The arrangement kept the two enzymes close together. Their laboratory system achieved complete depolymerisation of the tested PET film at 30°C, producing its constituent monomers without accumulated intermediate products.[5]
The study also examined reusing the cells, making the catalyst’s organisation part of the engineering problem. A useful recycling system needs the enzyme to keep working and be recoverable, as well as to perform a promising reaction once. This research took an existing AI-assisted enzyme and investigated how to deploy it more effectively.[5]
A later redesign strengthened the same starting point
A study in the May 2026 issue of the Journal of Hazardous Materials added a structural link to FAST-PETase. The redesigned version nearly completely depolymerised untreated post-consumer PET film within three days at 50°C, with periodic enzyme replenishment. The researchers reported roughly twice the speed of FAST-PETase under that comparison.[6]
Their explanation was a more stable connection between parts of the enzyme, improving its thermal stability and interaction with the plastic. This was further rational protein engineering, building on the earlier machine-learning-assisted variant. The sequence of work shows how different methods can contribute useful improvements to the same biological tool.[6]
The useful result is material changing form: old packaging supplies ingredients for a new polymer. FAST-PETase provides one part of that process. Longer-lived enzymes and better paired reactions offer ways to make more of the starting material available for reuse. The improvements can accumulate around a biological tool that already performs the central chemical task.
Sources & further reading
- Machine learning-aided engineering of hydrolases for PET depolymerization Nature · 27 Apr 2022 · Accessed 11 Oct 2026
- Plastic-eating enzyme could eliminate billions of tons of landfill waste The University of Texas at Austin · 27 Apr 2022 · Accessed 11 Oct 2026
- Discovery of novel gain-of-function mutations guided by structure-based deep learning ACS Synthetic Biology · Accessed 11 Oct 2026
- Supercomputers help reveal dynamic plastic-eating duo Texas Advanced Computing Center · 21 Jul 2021 · Accessed 11 Oct 2026
- Complete enzymatic depolymerization of polyethylene terephthalate (PET) plastic using a Saccharomyces cerevisiae-based whole-cell biocatalyst Environmental Science & Technology Letters · 19 Mar 2025 · Accessed 11 Oct 2026
- Rational redesign of FAST-PETase via a “locking” strategy for efficient PET depolymerization Journal of Hazardous Materials · Accessed 11 Oct 2026