We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Predicting molecular transformations of PBAT-derived dissolved organic matter in chlorobenzene-polluted system waters: A machine learning framework for sustainable pollution risk assessment
Summary
When "biodegradable" plastics (like PBAT, used in compostable bags and packaging) break down in water, they release dissolved substances that react more aggressively with pollutants like chlorobenzene than natural organic matter does—meaning these supposedly eco-friendly plastics may actually help toxic chemicals spread or transform in unexpected ways. Using machine learning, researchers found they could predict this reactivity with over 98% accuracy based on simple chemical traits, offering a tool to flag which plastic byproducts pose the highest pollution risks. This matters because it suggests biodegradable plastics aren't automat
Microplastic pollution poses significant challenges to cleaner production and sustainable water management. Dissolved organic matter leached from PBAT microplastics (PDOM) interacts with organic pollutants, altering their fate and carbon cycling, but the underlying molecular mechanisms remain unclear, limiting ecological risk prediction. This study innovatively integrates molecular network analysis with interpretable machine learning to systematically reveal the molecular transformation mechanisms and reaction patterns in the interaction between PDOM and a typical organic pollutant (chlorobenzene). Research reveals that compared to natural organic matter (NOM), PDOM exhibits more pronounced aromatic transformations and oxidation-related component conversions during binding processes. Molecular edges and average degree decreased by 28.67%, indicating a more reactive and dynamic molecular structure. We developed a quantitative analytical framework based on reaction networks, discovering that the number of demethylation and dehydrogenation reactions occurring in PDOM exceeds that in NOM by more than 5 times. Furthermore, we developed an interpretable machine learning predictive model that successfully identified the H/C ratio and molecular weight as key indicators for predicting the binding reactivity of PDOM and NOM molecules under pollutant stress, achieving > 98% prediction accuracy (the test set accuracy of the LightGBM model). This integrated framework enables quantitative prediction of DOM-pollutant interactions. The identified reactivity indicators provide a scientific basis for screening high-risk PDOM components and developing targeted pollution prevention strategies. Our findings directly support incorporating molecular-level DOM reactivity into environmental risk assessment, advancing cleaner production goals and sustainable management of microplastic-polluted aquatic ecosystems.