Molecular insights into potential hydrophobic deep eutectic solvents for furfural extraction guided by COSMO-RS and machine learning

Abstract

In this study, we aimed to gain molecular insights into the potential of hydrophobic deep eutectic solvents (HPDESs) as solvents for furfural extraction. To achieve this, we screened 108 HPDES constituents using the quantum chemical COSMO-RS method, saving time and resources. Moreover, the σ-profiles of the HPDES constituents were analyzed using machine learning techniques, including decision trees and multilinear regression, to understand their impact on extraction performance. Our results showed that the top 5 HPDES constituents were 3–5-di-tertbutylcatechol, 1-naphthol, p-hydroxybiphenyl, phenol, and thymol, indicating that aromatic-based solvents containing a phenolic hydroxyl group are ideal for furfural extraction. Building on this, we used the COSMO-RS method to predict the eutectic composition and melting temperatures of the top 10 binary HPDESs, composed of the top 5 pre-screened constituents. These HPDESs were then evaluated for furfural extraction under typical hemicellulose treatment conditions. Our findings showed that these HPDESs were highly effective for furfural extraction, achieving performance comparable to industry benchmarks while avoiding the high volatility associated with these benchmarks.