Revolutionizing inverse design of ionic liquids through the multi-property prediction of over 300,000 novel variants using ensemble deep learning

Abstract

In the flourishing field of materials science and engineering, ionic liquids (ILs) stand out for their advantageous features, unique tunable properties, and environmentally friendly attributes, making them ideal candidates for various applications. However, the enormous diversity of ILs presents a challenge that has traditionally been addressed through extensive experimental work. In this study, a computational approach that combines robust molecular modeling and advanced ensemble deep learning is employed. This proof-of-concept approach allows for the simultaneous prediction of multiple properties of ILs, thereby enabling a simplified pathway to eco-efficient inverse solvent design. Based on an extensive dataset from ILThermo with 73,847 data points of 2917 ILs from 1213 references and using insightful molecular features derived from COSMO-RS, 8 machine learning algorithms were used to predict various physical properties of ILs. Artificial Neural Networks (ANNs) have been proven to be the optimal choice based on the results obtained. The ANN model was carefully tuned, resulting in an ensemble model with a total of 11,241 parameters that exhibited remarkable predictive ability with R² values of 0.993, 0.907, 0.931, and 0.875 for density, viscosity, surface tension, and melting temperature, respectively. A remarkable feature of this study is the extensive screening of 303,880 ILs obtained by combining all possible pairs from a set of 1070 cations and 284 anions (1070×284). This demonstrates a pragmatic approach to identifying different property profiles that significantly narrow the spectrum for experimental validation. Based on the screening, an open-source “Inverse Designer Tool” was developed as an advanced database filter to explore ILs based on user-defined criteria, facilitating the identification of promising IL candidates for specific applications. The results presented here open a door for a new approach to the exploration and application of ILs and catalyze their integration in various industrial fields as potential environmentally friendly solvents.