Plate Nº 11 · recorded October 2, 2026
Biology & EvolutionReported finding
AI Searches Millions of Enzymes to Find Pollution Fighters
Murdoch University researchers combine machine learning and biochemistry to mine millions of natural enzymes for ones that can break down plastic, PFAS and other persistent pollutants already contaminating soils and crops.
By Nathan Brooks3 min read621 words
In brief
- Agricultural soils contain roughly 23 times more microplastics than oceans, according to a prior review by Joseph Boctor.
- Machine learning pipelines trained on characterized enzymes predict how untested enzyme structures might interact with and break down target pollutants.
- The review appears in Nature Reviews Earth & Environment (2026), DOI: 10.1038/s43017-026-00839-2.
Databases around the world hold records on millions of enzymes, many of which could potentially degrade pollutants such as plastic. Scientists at Murdoch University's Bioplastics Innovation Hub in Australia are now combining machine learning with biochemistry to work out which of these enzymes can actually break down plastic and other harmful contaminants.
Ph.D. candidate Joseph Boctor argues in a review published in Nature Reviews Earth & Environment that the key to matching the right biological tool to each pollutant likely already sits inside existing data. Sorting through it by hand would once have been an impossible task. AI technology is now filling that gap at a critical moment for the environment.
Mining evolution's existing solutions
"I strongly advocate that overengineering enzymes is a bad starting point that overlooks millions of years of evolution that have already produced lots of potential solutions to these contaminants," Boctor said.
"We need to look in the right place, and by leveraging machine learning tools, we are able to mine through millions of pieces of unexplored biological data to find the right candidate for the relevant task."
The idea is simpler than it sounds. Enzymes are proteins that catalyze chemical reactions in living organisms — in this case, reactions that can snip apart the molecules making up pollutants. The machine learning pipelines Boctor describes train on enzymes that scientists have already characterized in the lab. From that knowledge, the models predict how the structures of untested enzymes might interact with a target pollutant and break it down.
Pollutants moving from soil to food
Boctor stresses that while researchers work on bioplastic alternatives for the future, these computational tools help tackle the pollutants already circulating in the environment and causing active harm.
"PFAS, microplastics and other persistent pollutants are not only industrially favorable but biologically active," he said. "They trick our bodies by mimicking our own hormones and causing documented disruptions to health."
PFAS, often called "forever chemicals," are a family of synthetic compounds widely used in industry that degrade extremely slowly. Their persistence is exactly what makes them so difficult to remove from soil and water.
The urgency is grounded in Boctor's own earlier findings. Last year, he conducted a comprehensive review showing that agricultural soils contained around 23 times more microplastics than the oceans. That review also reported detections of microplastics and nanoplastics — plastic fragments down to microscopic scale — in lettuce, wheat and carrot crops.
Soil additives raise additional concerns. Phthalates, chemicals used to soften plastics, have been linked to reproductive issues. PBDEs, a group of flame retardants, are neurotoxic and associated with neurodegenerative disease, increased risks of stroke and heart attack, and early death.
With these pollutants traveling from soil to salad to the human body, the search for enzymes that can break them down has taken on new importance.
From prediction to validation
Machine learning saves researchers the enormous time once required to search through existing data by hand. But prediction alone does not clean up a contaminated field. The Bioplastics Innovation Hub team is now directing its energy toward testing, validating and scaling the enzymes the algorithms identify, with the goal of deploying them in bioremediation — the use of living organisms or their products to remove pollutants from the environment.
How quickly candidate enzymes can move from computational shortlist to real-world cleanup remains to be seen. The review outlines a promising direction rather than a finished solution, and the team's results in laboratory and field settings will determine whether evolution's archive truly holds the pollution fighters we need.
Publication details: Joseph Boctor, Using machine learning with biochemical analysis to identify suitable enzymes for bioremediation applications, Nature Reviews Earth & Environment (2026). DOI: 10.1038/s43017-026-00839-2.
via Phys.org Biology (Source)
More from Nathan Brooks
Nearby plates
- AI Scientist Makes Biological Discoveries With Minimal Human Help
- 392 Million Records: New Database Maps Europe's Changing Food Risks
- Why the Same Chemical Harms One Species and Spares Another
- Meteorite Sugar May Have Shaped Early Earth's Chemistry
- Ancient Ocean Chemistry Kept Earth Breathing, Study Finds