TP-Transformer: An Interpretable Model for Predicting the Transformation Pathways of Organic Pollutants in Chemical...

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Toxicology

Background:

• Chemical oxidation is crucial for removing organic pollutants from water.
• Oxidation can create harmful transformation products (TPs) that require identification for risk assessment.
• Current methods for identifying TPs are costly and time-consuming.

Purpose of the Study:

• To develop an advanced deep learning framework, TP-Transformer, for predicting the structures and formation pathways of TPs.
• To provide a scalable and efficient tool for environmental risk assessment and the optimization of oxidation processes.

Main Methods:

• Developed TP-Transformer, a deep learning model trained on the Chem_Oxi_2K dataset (2780 degradation reactions).
• Utilized attention mechanisms to analyze substrate reactivity and reaction conditions.
• Validated predictions through experimental testing on pollutants not included in the training data.

Main Results:

• TP-Transformer achieved 86.28% accuracy in predicting TPs on the training dataset.
• The model successfully elucidated complete degradation pathways.
• Experimental validation showed 80.20% to 92.86% accuracy for novel pollutants.

Conclusions:

• TP-Transformer offers a precise, efficient, and scalable method for identifying TPs and their pathways.
• The model has the potential to significantly advance water treatment strategies and protect environmental and human health.