Recent research published in the journal Engineering has unveiled a promising approach to designing deep eutectic solvents (DESs) using machine learning (ML). Conducted by Usman L. Abbas and his team at the University of Kentucky, this study addresses the ongoing challenge of discovering new DES candidates, which are non-ionic solvents with significant applications in various industries including catalysis, extraction, carbon capture, and pharmaceuticals.
Traditionally, the search for new DESs has relied on intuition and trial-and-error methods, leading to low success rates and missed opportunities. Recognizing the crucial role that hydrogen bonds (HBs) play in the formation of DESs, the researchers aimed to identify specific HB characteristics that differentiate DESs from non-DES systems. Their analysis involved studying the molecular dynamics (MD) simulation trajectories of 38 known DESs and 111 non-DES systems.
The findings revealed two key features that set DESs apart: they exhibit a greater imbalance between the numbers of intra-component HBs and possess more numerous and stronger inter-component HBs. These insights provided a foundation for developing 30 ML models that utilized various algorithms and HB-based descriptors to predict DES formation.
The researchers validated their models against experimental data from 34 systems, with the extra trees forest model achieving the highest performance, boasting an ROC-AUC score of 0.88. Abbas emphasized the significance of their findings, stating, “Our work illustrates the importance of HBs in DES formation and shows the potential of ML in discovering new DESs.”
The commercial implications of this research are substantial. Industries that rely on solvents for chemical processes could benefit from the ability to efficiently discover and develop new DESs tailored to specific applications. For example, the pharmaceutical sector could leverage these insights to create more effective drug formulations, while the carbon capture industry might find novel solvents that enhance the efficiency of CO2 absorption. Additionally, the extraction industry could utilize advanced DESs for improved separation processes.
As the demand for sustainable and efficient solvents continues to grow, the integration of machine learning in solvent design presents a transformative opportunity. This research not only paves the way for innovative solutions but also highlights the intersection of chemistry and technology, showcasing how modern methodologies can enhance traditional practices across various sectors.
