Machine Learning-Enhanced DGT Passive Sampling Coupled with Non-Targeted Analysis for High-Throughput Monitoring of...

Area of Science:

• Environmental Chemistry
• Analytical Chemistry
• Computational Chemistry

Background:

• High-throughput monitoring of aquatic organic pollutants is challenging due to complex sources and lack of standards.
• Diffusive Gradients in Thin-films (DGT) offers time-integrated sampling, while Nontargeted Analysis (NTA) provides broad chemical identification.
• DGT-NTA is limited by the need for pollutant-specific diffusion coefficients (D) and ionization efficiencies (IE) for quantification.

Purpose of the Study:

• To develop a machine learning (ML)-assisted framework for semiquantitative analysis in DGT-NTA.
• To predict D and IE values, reducing the need for compound-specific standards.
• To enable high-throughput environmental monitoring of organic pollutants.

Main Methods:

• Developed ML models to predict diffusion coefficients (D) and ionization efficiencies (IE).
• Applied the ML-assisted framework to DGT-NTA data from municipal wastewater effluents.
• Validated the predicted concentrations against targeted analysis.

Main Results:

• Successfully enabled semiquantitative estimation of 85 identified pollutants in wastewater effluents.
• Achieved a mean prediction error of 1.78-fold for compounds within the ML models' applicability domains.
• Prioritized 8 potentially high-risk pollutants through ecological risk assessment.

Conclusions:

• The ML-assisted framework significantly reduces the dependence on analytical standards in DGT-NTA.
• This approach offers a promising solution for high-throughput monitoring of emerging pollutants in aquatic environments.
• Facilitates ecological risk assessment by providing semiquantitative pollutant concentrations.