Machine Learning Reveals Key Adsorption Mechanisms for Oxyanions Based on Combination of Experimental and Published Literature Data

Abstract

The development of new adsorbents for water treatment often involves complex adsorption mechanisms, whose individual contributions are unclear, thereby limiting the understanding of adsorption driving forces, making it difficult to achieve precise design of adsorbents. Machine learning (ML) has been used to uncover the impacts of these mechanisms through feature engineering, but progress is limited by the data quality for training. Herein, we developed a universal ML strategy for precisely predicting the adsorption capacity of polymers for oxyanions and identifying the adsorption driving force based on the combination of experimental and published literature data. The adsorption mechanism was explored through classification of RDkit descriptors with different SHAP importance values, and electrostatic interaction was found to be the driving force in the oxyanion adsorption process, which was further verified by theoretical calculations, adsorption experiments, and effective targeted adsorbent design. In comparison, analysis relying on a separate literature data source led to decreased model performance, some biased conclusions, and invalid targeted adsorbent design. Overall, this study proposed a strategy for data set optimization as well as dominant mechanism identification, which could shed light on better treatment of oxyanions in wastewater.