Machine learning for ionic liquid discovery in gas absorption: From solubility modeling to high‐throughput screening
TL;DR
This work presents a comprehensive data‐driven modeling framework to predict the solubility of ILs for 10 gases of broad interest in the literature, establishing promising predictive models for multi‐gas solubility in ILs, accelerating the selection of ILs for gas‐involved applications.
Abstract
Gas solubility is critical for designing ionic liquids (ILs) in gas‐related applications, yet the vast chemical and operational space renders experimental exploration challenging. This work presents a comprehensive data‐driven modeling framework to predict the solubility of ILs for 10 gases of broad interest in the literature. Five molecular representations and five machine/deep learning models are systematically evaluated under both data point‐based and IL‐based dataset splitting strategies. By leveraging the pre‐trained model obtained from CO 2 solubility, the transfer learning strategy is explored to further improve the model performance for the other nine gases. Model interpretability is elucidated via SHAP analysis and attention mechanisms, and the practical utility of developed models is demonstrated through high‐throughput screening of more than 100 million potential ILs (the largest dataset reported to date). In summary, this work establishes promising predictive models for multi‐gas solubility in ILs, accelerating the selection of ILs for gas‐involved applications.