Predicting deep eutectic solvents for absorption of SO 2 based on multilayer perceptron

Sulfur dioxide (SO 2 ) is a common air pollutant, primarily emitted from fossil fuel combustion and industrial flue gas. Deep eutectic solvents (DESs) have been proven to be innovative, sustainable SO 2 absorbents. Meanwhile, the relationships between DESs molecules and their SO 2 capacity or SO 2 absorption selectivity in the gas source containing carbon dioxide (CO 2 ) are unclear. In this work, a multilayer perceptron (MLP) model was developed to predict the SO 2 capacity of DESs based on the database containing 1382 SO 2 capacity data from the literature. The SO 2 capacity with respect to the chemical and physical absorption was investigated. The optimal MLP model (R 2 (coefficient of determination) = 0.9924, MSE (mean square error) = 0.0009) with 13 input descriptors including COSMO-RS (Conductor-like Screening Model for Realistic Solvents) theory descriptors and temperature, pressure, and water concentration validated its superior predictive capability in forecasting both physical and chemical absorption. From the SHAP (Shapley Additive Explanations) analysis, the DESs candidates with high SO 2 capacity should consider anionic and relatively long-branched components. A database containing 924 novel DESs was constructed. Through the integration of this model with a predictive model for CO 2 capacity, the promising candidates with high selectivity of SO 2 , such as [N444]Cl (tetrabutylammonium chloride) and 2-Mim (2-methylimidazole), were effectively identified within the novel database. This work offers insights for screening suitable DESs for the flue gas desulphurization (FGD) process.