Extension of scoring-assisted generative exploration for ionic liquids (SAGE-IL) and its application to ionic liquid design for CO2 capture

Greenhouse gas emissions, particularly carbon dioxide (CO 2 ), pose a significant threat to the climate, driving global warming and numerous environmental issues. Ionic liquids (ILs) have gained attention for CO 2 capture due to their tunable properties, which can be optimized through the selection of specific cation-anion combinations. Traditional methods for discovering effective ILs are time-consuming and costly, necessitating the development of innovative computational approaches. This study extended the Scoring-Assisted Generative Exploration for Ionic Liquids (SAGE-IL) to generate novel ILs tailored for CO 2 capture. SAGE-IL combines deep learning-based generative models with quantitative structure-property relationship models to design ILs with desired properties. Notably, SAGE-IL successfully performed single property optimization, either increasing or decreasing viscosity and melting point, through iterative interactions between generative and evaluation models. Furthermore, it achieved multiple property optimization, improving either CO 2 solubility or activity coefficient while simultaneously optimizing density, melting point, viscosity, and synthetic accessibility. This approach not only accelerates the discovery of high-performance ILs but also highlights the potential of generative models in de novo molecular design, paving the way for future advancements in materials discovery.