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

• DOI: 10.1016/j.mser.2024.100798
• Publication Date: 2024-05-07T21:30:33Z
• Authors: Tarek Lemaoui; Tarek Eid; Ahmad S. Darwish; Hassan A. Arafat; Fawzi Banat; Inas AlNashef
• Keywords: Ensemble Learning

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