Predicting Adsorption of Volatile Organic Compounds onto Biochars with Machine Learning and Potential Applications

Volatile organic compounds (VOCs) are a significant health and environmental concern due to their potential toxicity to human beings and their detrimental role in atmospheric photochemical reactions. While VOC removal by biochar offers a promising strategy for VOC management, enhanced adsorption performance using the intricate interactions between VOCs and biochar has not been fully elucidated. This study introduced a sophisticated and accurate machine learning (ML) model trained from an automated ML frame using a comprehensive data set encompassing 52 biochar materials with varied characteristics (elemental composition and physical properties), 15 representative VOCs of different molecular properties (manifested by RDKit molecular fingerprint descriptors), diverse adsorption conditions (temperature, pressure, dose of biochar, and initial concentration of VOCs), and the corresponding VOC removal capacities. The model achieved an impressive test R2 of 0.97 and a root-mean-square error (RMSE) of 1.22 mmol/g, underscoring its robustness. The feature importance analysis highlighted the dominance of physical adsorption in VOC removal capacity under conditions of high surface area and pore volume, particularly at high pressures for biochar with a high micropore volume. The number of hydroxyl groups (fr_Al_OH) of VOCs was identified as a crucial molecular descriptor, representing that VOCs with high polarity can positively influence biochar adsorption. Conversely, the topology size (BCUT2D_MRHI) of VOCs was negatively correlated with the adsorption capacity, indicating the preference for biochar to remove smaller VOC molecules. The broad application domains of our model, defined by kernel density estimation, demonstrated a high value in the evaluation of VOC-adsorption biochar materials, effectively serving the management and control of emerging contaminants in VOCs.