Machine learning for deep eutectic solvents: advances in property prediction and molecular design

Deep Eutectic Solvents (DES) have gained attention as eco-friendly alternatives to standard solvents, providing adaptability, biodegradability, and cost-effectiveness for use in drug delivery, catalysis, and energy storage applications. However, their complicated molecular interplay and compositional fluidity make experimentally determining properties such as viscosity, density, and melting point a significant undertaking. ML has increasingly been utilized as a powerful approach for predicting DES properties by identifying nonlinear relationships between molecular descriptors and macroscopic behaviors. This survey offers an extensive examination of ML applications in DES property prediction, covering prominent algorithms like artificial neural networks, support vector machines, random forests, and XGBoost, characteristic engineering strategies such as COSMO-RS descriptors and SMILES encoding, as well as hybrid methods merging physics-based modeling. We critically assess model performance metrics including R2, RMSE, and MAE, dataset qualities, and validation protocols, highlighting the superiority of ensemble and physics-informed ML methods in accuracy and generalizability. Recent advances in ML, for instance multifidelity Gaussian Process Regression and graph neural networks, demonstrate enhanced predictive capabilities for viscosity, density, and melting points, surpassing traditional thermodynamic models such as COSMO-RS and UNIFAC. Interpretability tools like SHAP analysis uncover dominant molecular interactions, guiding DES structure for specific applications. Challenges like data scarcity, extrapolation limitations, and computational expenses are discussed, alongside solutions like transfer learning and active learning. The study also explores burgeoning directions, involving the integration of quantum chemical descriptors and autonomous high-throughput screening. By bridging computational intelligence with chemical insights, ML accelerates the discovery of customized DESs, fostering innovation in green chemistry.