Machine Learning Approach for the Prediction of Eutectic Temperatures for Metal-Free Deep Eutectic Solvents

Abstract

Deep eutectic solvents (DESs) represent an environmentally friendly alternative to conventional organic solvents. Their liquid range determines the areas of application, and therefore, the prediction of solid-liquid equilibrium (SLE) diagrams is essential for developing new DESs. Such predictions are not yet possible by using the current state-of-the-art computational models. Herein, we present an alternative model based on support vector regression integrating experimental data, a conductor-like screening model for real solvents simulations, and cheminformatic descriptors for predicting melting temperatures of binary metal-free DESs or ionic liquids, allowing the researcher to estimate the eutectic formation and SLE for specific combinations of components. The model was developed based on the manually collected database of 1648 mixture melting temperatures for 237 experimentally described DESs, and its accuracy was demonstrated by 5-fold cross-validation (R2 ∼ 0.8). The presented machine learning methodology empowers researchers to predefine the liquid range of the mixture and holds promise for efficient molecular combination screening, facilitating the discovery of tailored DESs for desired applications from catalysis and extraction to energy storage. By enabling a deeper understanding of DES behavior and the targeted design of these solvents, the proposed approach contributes to advancing green chemistry practices and to promoting sustainable solvent usage.