We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Discovering hidden candidate plastic-degrading enzymes: Combined multi-omics and machine learning strategy
Summary
Scientists used AI and genetic analysis to hunt for natural bacterial enzymes that can break down common plastics like polyethylene and polyurethane, materials linked to microplastic pollution that ends up in our food, water, and even bloodstream. By studying microbes that had been feeding on plastic for a long time, they identified promising new plastic-eating enzymes that hadn't been found before, some made by a bacterium called Rhodococcus aetherivorans. This matters because finding better plastic-degrading enzymes could eventually lead to more effective ways to clean up plastic waste, potentially reducing the microplastics
Plastic pollution poses a major threat to the stability of natural ecosystems as well as human health. Microbial enzymes have long been considered a potential resource for targeted biodegradation but, except for a few successful cases, the discovery of efficient enzymes has proved challenging. Aiming to accelerate the process, we propose an approach combining metagenomics, metatranscriptomics and semi-supervised learning that selects promising plastic-degrading candidate enzymes from the proteome of relevant microorganisms. Tested on a dataset of over 10,000 microbial proteins, ranking models consistently prioritize known plastic-degrading enzymes, achieving an area under the cumulative distribution function curve above 0.96, with leave-one-family-out cross-validation indicating that performance is largely retained across protein families. As a case study, this work focuses on mixed microbial cultures exposed for extended periods to polyethylene, polyethylene terephthalate, and polyurethane substrates. The prevalent species after selective enrichment were functionally characterized, finding Rhodococcus aetherivorans as the most relevant species in two of the five cultures under investigation. Among the top-ranked proteins, several have high structural similarity with known enzymes despite not being identified by sequence similarity search. Moreover, according to metatranscriptomics results, several of these enzymes were found to be expressed at the same level or above that of annotated enzymes, suggesting that they may have functional relevance. Overall, this work highlights the potential of integrating multi-omics with data-driven methods for enzyme discovery and for accelerating the development of biotechnological solutions to plastic pollution.