Artificial Intelligence-Powered Prediction of Microplastic-Pollutant Adsorption Coefficients Enables Scalable Risk...

Area of Science:

• Environmental Science
• Computational Chemistry
• Ecotoxicology

Background:

• Microplastics (MPs) are ubiquitous environmental contaminants.
• MPs can adsorb hazardous organic pollutants, influencing their environmental fate and transport.
• Accurate prediction of microplastic-pollutant adsorption is crucial for environmental risk assessment.

Purpose of the Study:

• To develop a predictive model for microplastic-pollutant adsorption.
• To enhance the characterization of sorption behavior using multimodal data.
• To provide a tool for high-throughput prediction across various environmental conditions.

Main Methods:

• Developed a multimodal Siamese neural network (MPAP).
• Integrated molecular fingerprints, graph embeddings, MP morphology, and water chemistry data.
• Trained the model on 1101 adsorption records (403 compounds, 6 MP types).
• Validated predictions with experimental batch adsorption and mass spectrometry.

Main Results:

• The MPAP model achieved high accuracy (R² = 0.869 validation, R² = 0.863 test).
• Experimental validation confirmed the model's strong predictive performance.
• The model effectively characterizes microplastic-pollutant sorption behavior.

Conclusions:

• The MPAP model offers a robust and accurate method for predicting microplastic-pollutant adsorption.
• The multimodal approach provides a comprehensive understanding of sorption dynamics.
• An open-access web platform facilitates broad application and research.