How is machine learning used in environment?
Pusan National University Researchers Innovate Environmental Health with Advanced Machine Learning Technique
Researchers from Pusan National University have made significant strides in environmental health by introducing an innovative machine learning approach called Feature Localized Intercept Transformed-Shapley Additive Explanations (FLIT-SHAP). This groundbreaking method has unveiled crucial interactions among pollutants, significantly improving toxicity predictions and aiding in environmental health decision-making.
Enhancing Toxicity Predictions with FLIT-SHAP
Traditionally, environmental health research has focused on the toxicity of individual pollutants, often overlooking the complex interactions that occur when multiple pollutants are present. This new study, however, highlights the importance of understanding these interactions. Led by Professor Kuk Cho, the research team developed FLIT-SHAP to better predict the oxidative potential (OP) of PM2.5, a type of particulate matter known to cause various health issues.
FLIT-SHAP stands out because it provides detailed insights into the effects of specific pollutants within a mixture, unlike traditional methods that treat pollutant effects more broadly. This precise approach is particularly beneficial for analyzing the OP of particulate matter in the air. The study’s findings were published in the Journal of Hazardous Materials on August 14, 2024.
Understanding Oxidative Potential
“The OP of PM refers to the ability of these particles to generate reactive oxygen species, which can cause oxidative stress and inflammation in the body, leading to various health problems such as respiratory and cardiovascular diseases,” explained Prof. Cho. By employing FLIT-SHAP, the researchers could examine the OP of these particles within a mixture of pollutants.
Significant Findings and Real-World Applications
Using laboratory-controlled data, the team trained robust machine learning models, including eXtreme Gradient Boosting (XGBoost). FLIT-SHAP revealed significant insights, demonstrating that 55-63% of interactions in a controlled environment were synergistic, increasing toxicity, while 25-42% were antagonistic, reducing toxicity. In real-world scenarios, however, antagonistic effects were more prevalent, leading to lower overall toxicity than predicted by traditional models.
One of the most surprising discoveries was the significant impact of quinones, a group of chemicals previously underestimated in their contribution to toxicity. This finding challenges current regulatory approaches and suggests a need for greater focus on these chemicals.
Implications for Environmental Health and Regulation
FLIT-SHAP not only provided highly accurate toxicity predictions (R² = 0.99) but also offered detailed insights into pollutant interactions. This makes it a powerful tool for conducting realistic risk assessments and enhancing our understanding of environmental health risks. The researchers’ rigorous tests revealed that traditional additive models often misjudge the toxicity of pollutant mixtures, emphasizing the need for more refined risk assessments.
While the study focused on a select group of pollutants and toxicity endpoints, further research is needed to explore broader scenarios. Nonetheless, FLIT-SHAP represents a significant leap forward, offering a precise tool for evaluating chemical mixture toxicity. This advancement promises to improve regulatory decisions, ultimately safeguarding human health more effectively.
The introduction of FLIT-SHAP by Pusan National University researchers marks a pivotal development in environmental health research. By providing a detailed analysis of pollutant interactions, this innovative machine learning approach enhances toxicity predictions and supports better decision-making processes in environmental health. As further research continues, FLIT-SHAP has the potential to revolutionize how we assess and regulate environmental pollutants, ensuring a healthier future for all.
Reference
Title of original paper: Machine learning-derived dose-response relationships considering interactions in mixtures: Applications to the oxidative potential of particulate matter
Journal: Journal of Hazardous Materials
DOI: https://doi.org/10.1016/j.jhazmat.2024.134864
