Rapidly Predicting Aqueous Adsorption Constants of Organic Pollutants onto Polyethylene Microplastics by Combining Molecular Dynamics Simulations and Machine Learning

Abstract

Adsorption of aqueous organic pollutants onto microplastics influences the exposure and risks of both the pollutants and microplastics. Experimental determination of the aqueous adsorption equilibrium constants (K_aq) that characterize the adsorption capacity of microplastics to pollutants is laborious and inefficient since the K_aq values rely on various combinations of conditions, such as pH, ionic strength, and particle sizes. Herein, molecular dynamics (MD) methods were established by comparing the MD-calculated K_aq values with the empirical values of 14 compounds adsorbed onto polyethylene (PE) microplastics having different particle sizes (10–250 μm) in pure water and seawater. Based on the data sets consisting of experimental and MD-calculated K_aq values, machine learning models were constructed. A gradient boosting decision tree (GBDT) model requires only easily obtainable Mordred descriptors for pollutants and desired conditions (particle sizes and ionic strength) to yield accurate results, with an external determination coefficient of 0.99. The GBDT model exhibits a great improvement over the previous one, as it incorporates multiple factors including ionic strength from pure water to seawater, dissociation species at different pH, and PE particle sizes with diameters ranging from nanometers to micrometers. This study paves a new way for high-throughput estimating K values for microplastics and pollutants at different environmental conditions.