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

Models to predict the environmental fate of micropollutants are needed for alternatives assessment and safe-by-design efforts. Wastewater treatment plants (WWTPs) are the main barrier to prevent micropollutants from entering receiving water bodies, and WWTP breakthrough is an important indicator of chemical persistence. State-of-the-art models to predict breakthrough are limited by their need for first-order degradation rate constants, a metric that is often unavailable. Here, we build models that predict removal in conventional treatment directly from the chemical structure using data from field-scale monitoring for over 1000 chemicals. The best predictions were achieved using substructure-based fingerprints (i.e., MACCS) and random forests, and identified influential substructures agree with structural moieties relevant for biotransformation. We show that our models are more reliable than existing process-based models used in EU and US regulatory contexts, making them important contributions to the in silico toolbox for alternatives assessment, the design of more benign chemicals in industrial research and development, and even exposure modeling in a risk assessment context. Moreover, our data sets along with our extensive systematic evaluation of different curation criteria and the scripts to reproduce it are key for future model advancement. Our model is publicly available (pepper-app) along with the training data and the scripts to reproduce the data curation process (github.com/FennerLabs/pepper).

Micropollutants are often only partially removed in conventional wastewater treatment, so models to accurately predict the level of breakthrough are desired but lacking. We present statistical models based on monitoring data as a novel avenue to tackle this problem.