Machine Learning Prediction of Melting Points in Nonionic Deep Eutectic Solvents: A Computational Framework Integrating Molecular Descriptors and Physicochemical Insights

Deep eutectic solvents (DES) have emerged as sustainable alternatives to conventional organic solvents, yet rational design remains hindered by limited predictive tools. Machine learning models are developed here to predict melting points across 1824 experimental measurements spanning 139 binary nonionic DES systems. The XGBoost model achieved robust predictive performance (R2 = 0. 909, RMSE = 17. 8 K, MAE = 10. 8 K) under rigorous mixture-based cross-validation. SHAP analysis revealed that the charge-weighted molecular surface area (PEOEVSA1) dominates melting point behavior, followed by pure component melting temperatures. Mixture-weighted descriptors capturing nonadditive interactions account for 58% of total feature importance, underscoring synergistic HBA–HBD effects. Hydrogen bond donor properties exert greater influence than acceptor characteristics (16. 6 vs 9. 9%). The model enables rapid virtual screening of candidate DES formulations, reducing experimental burden by 99% while maintaining interpretability through feature importance analysis.