Machine learning-assisted separation of lignin from hydrothermal treatment of biomass by deep eutectic solvents

Pretreatment of biomass using deep eutectic solvents (DESs) is a promising method for lignin separation. However, the properties of the separated lignin depend on labor-intensive experiments, lacking rational design and optimization. This study integrates machine learning with DES pretreatment to predict lignin yield, purity, weight-average molecular weight (Mw), and number-average molecular weight (Mn). The optimized gradient boosting regression models demonstrated high predictive accuracy and generalization with R² of 0.95-0.98, 0.78-0.97, and 0.82-0.90 in the training, testing, and validation datasets, respectively. Feature analysis revealed that reaction conditions (e.g., temperature and time) critically influenced lignin yield and purity, whereas DES characteristics primarily determine the Mw and Mn. Appropriately increasing temperature and time enhanced yield and purity, while a higher initial lignin content or surface area of DES was essential for obtaining high-molecular-weight lignin. These findings helped elucidate complex reaction mechanisms and improve the yield and quality of the separated lignin.