A machine-learning approach clarifies interactions between contaminants of emerging concern

Humans and biotas are exposed to a cocktail of contaminants of emerging concern (CECs), but mixture regulation is lagging behind. This is largely attributed to inadequate experimental data of mixture risk; revealing intricate interactions among CECs in mixtures with random combinations remains a formidable challenge. Here, we propose a new framework comprised of 5,720 lab tests of mixture risk for 100 CECs with random combinations, extended prediction of mixture risk in any CEC combination via a new machine learning model, and validation in field sites. We identify a general concave-down relationship between CEC number and ecological risk of algae, invertebrates, and fish under different lab conditions and in more than 900 field sites worldwide. We propose a new redundancy mechanism'' to clarify interactions among CECs, suggesting implications in grouping CECs by action mode for developing mixture regulatory frameworks. Our framework provides a blueprint for addressing cocktail effects of multi-factors with random combinations in different disciplines.

Automated summary

In this study, a new framework is proposed to address the mixture regulation of emerging contaminants of concern (CECs). The framework includes experimental data, a machine learning model, and field site validation to reveal the relationship between CEC number and ecological risk. The study also introduces a new redundancy mechanism to clarify interactions among CECs.