Screening High-Efficiency Deep Eutectic Solvents for Oxidative-Extractive Desulfurization via a Machine Learning Framework to Analyze Intermolecular Interactions

Oxidative-extractive desulfurization (OEDS) represents a highly efficient pathway for removing recalcitrant thiophenic sulfides from liquid fuels. Deep eutectic solvents (DESs) have emerged as promising green extractants due to their tunable multi-site non-covalent interactions and favorable environmental profile. However, exploring the vast chemical space of potential DES formulations demands rational, data-driven methodologies to identify molecular affinities while bypassing costly experimental trial-and-error. Herein, we report a novel hybrid algorithm termed PDI-DMPNN+SME. By coupling solvation free energy predictions with substructural masking, this framework quantitatively decodes the affinity and repulsion aimed at target molecules. The algorithm systematically evaluates both the extraction affinity of candidate DESs toward oxidized thiophenic species and the internal eutectic feasibility between solvent components. Through high-throughput screening of 19,291 binary candidates, an environmentally benign citric acid-glycerol DES system was identified as a promising extractant candidate. Under optimized process conditions, this system achieved a single-stage dibenzothiophene (DBT) removal efficiency of 99.95% in model oil and 93.72% in real diesel, maintaining satisfactory performance (98.50% and 90.80%, respectively) after the tested three consecutive regeneration cycles. Density functional theory (DFT) calculations reveal that the total interaction energy between the citric acid-glycerol system and dibenzothiophene sulfone (DBTO2) reaches −31.24 kcal/mol. Energy decomposition analysis indicates that electrostatic and dispersion interactions contribute comparably to this stabilization, jointly overcoming Pauli repulsion to form a thermodynamically favorable extractant-solute complex. This work establishes an intelligent, green paradigm for the rational design of functional solvents in advanced separation processes.