Deciphering the adsorption mechanisms between microplastics and antibiotics: A tree-based stacking machine learning approach

Water pollution has long been a focal point of environmental research, with its significance expected to persist in the foreseeable future. The emergence of the major pollutants has brought heightened attention to the contamination of water bodies by microplastics (MPs) and antibiotics (ABX), two prevalent contaminants that pose a growing concern. The intrusion of MPs into biological chains and the proliferation of antibiotic resistance genes (ARGs), primarily driven by ABX, underscore the synergistic amplification of pollution when these contaminants interact. This dynamic interplay accentuates the critical need to elucidate the adsorption mechanisms, the principal mode of interaction between MPs and ABX. While substantial research has been conducted on prevalent MPs and ABXs, the heterogeneity of these pollutants necessitates a shift towards developing robust adsorption models that can efficiently integrate diverse experimental data. Traditional isothermal adsorption models used in previous studies are limited in their generalizability, often considering only a single adsorbent or adsorbate under restricted experimental conditions and environmental parameters. Recently, machine learning (ML) models have shown promise in overcoming these limitations. With appropriate feature engineering, ML models can demonstrate remarkable generalization capabilities and provide a comprehensive understanding of the underlying mechanistic explanations. In this study, we developed and trained an ML model using 303 datasets sourced from open-access scientific literature. After comparing various single, composite, and deep learning algorithms, the tree-based Stacking model demonstrated superior predictive performance, achieving an R2 of 0.91 and an RMSE of 0.12. To address the common "black-box" issue of ML models, we employed SHapley Additive exPlanations (SHAP) to interpret the model's predictions, elucidating the interaction mechanisms between MPs and ABX. Our results reveal that chemical factors such as hydrogen bond formation, it-it interactions, and the presence of oxygen-containing functional groups significantly influence the adsorption process. Additionally, physical factors like van der Waals forces on molecular surfaces play a crucial role. The stacking model not only accurately predicts adsorption coefficients across a wide range of MPs and ABX but also strengthens our understanding of the adsorption mechanisms.