Ionic liquids (ILs) are promising electrolytes for sustainable energy storage devices owing to their low volatility, nonflammability, and excellent electrochemical stability. However, the data scarcity of IL has hindered the rational design for diverse applications. Here, we present a BRICS-based approach to construct the largest IL database to date, comprising 167,350 cations, 2460 anions, and nearly 108 potential ILs. According to the high-throughput DFT calculation and literature search, we accumulate ∼105 labeled properties for ILs in the database, which lays the foundation to develop AI models in predicting the ionic conductivity, electrochemical window, and melting point of the ILs for ∼107 generated ILs. We reveal that XGBoost and MLP exhibit optimal performance even with small data sets, whereas graph-based models, such as GCN and GAT, require larger data sets to achieve similar accuracy. We introduce an ILScore to screen high-performance ILs with synthetic feasibility and validate it using molecular dynamics simulations of representative 200 candidates for lithium battery electrolytes. A transport decoupling factor (TDF) has been proposed to distinguish three lithium conduction mechanisms in the IL electrolytes. This work illustrates a chemistry-informed AI and computational approach to overcome data scarcity in IL research while also supplying a rich data set for future IL design and property-prediction initiatives.
Wang et al. (2025) studied this question.