Predicting Micropollutant Removal in Wastewater Treatment Based on Molecular Structure: Benchmark Data and Models

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Chemical Engineering

Background:

• Accurate prediction of micropollutant environmental fate is crucial for chemical safety assessments and designing safer chemicals.
• Wastewater treatment plants (WWTPs) are key in preventing micropollutant release, with breakthrough indicating persistence.
• Current models often require unavailable degradation rate constants, limiting their applicability.

Purpose of the Study:

• To develop predictive models for micropollutant removal during conventional wastewater treatment.
• To utilize chemical structure as input, bypassing the need for experimental degradation data.
• To provide a reliable *in silico* tool for regulatory and industrial applications.

Main Methods:

• Developed predictive models using machine learning algorithms, specifically random forests.
• Employed substructure-based fingerprints (e.g., MACCS) derived from chemical structures.
• Trained and validated models using field-scale monitoring data for over 1000 chemicals.

Main Results:

• Models based on MACCS fingerprints and random forests demonstrated high predictive accuracy for WWTP removal.
• Identified key substructures influencing removal align with known biotransformation pathways.
• The developed models outperformed existing process-based models used in regulatory frameworks.

Conclusions:

• The novel structure-based models offer a reliable and accessible method for predicting micropollutant WWTP breakthrough.
• These models enhance the *in silico* toolkit for alternatives assessment, safe-by-design initiatives, and risk assessment.
• Publicly available data, code, and the PEPPER model facilitate future advancements in environmental fate modeling.