Machine learning-driven QSAR study on efficient extractants for Mn2+-Mg2+ separation: Impact of steric hindrance and electronic characteristics on extraction performance

Manganese is a critical metal with widespread applications. The increasing demand for manganese, driven by the rapid growth of the electric vehicle and new energy sectors, has underscored the need for efficient separation and purification technologies, particularly from low-grade ores with high content of impurities. This study applies machine learning (ML) −driven quantitative structure–activity relationship (QSAR) modeling to investigate the Mn 2+ and Mg 2+ separation performance of 14 extractants. Key molecular descriptors were identified through principal component analysis and clustering heatmap method, and QSAR models were developed using ordinary least squares, nonlinear model, artificial neural network, random forest, and extreme gradient boosting (XGBoost) tree. The XGBoost model exhibited superior predictive accuracy for Mn 2+ and Mg 2+ extraction efficiencies and Mn 2+ /Mg 2+ separation selectivity. Feature importance analysis revealed that molecular properties related to hydrophilicity, electronic characteristics, and steric hindrance, as reflected by descriptors such as Clog P, OAV, Nonpolar S, VIP, Max P, and EI, play decisive roles in governing extraction efficiency and separation performance. Further NCI analysis demonstrated that the C–OH group of one α -hydroxyalkyl phosphinic acid ( α -HA) molecule forms a hydrogen bond with the C–OH group of another α -HA molecule, which likely accounts for the tendency of α -HA molecules to form dimers during solvent extraction. This interaction modulates the spatial arrangement of P–OH and P=O groups, enhancing both Mn 2+ extraction efficiency and Mn 2+ /Mg 2+ separation selectivity. These findings underscore the value of ML-driven QSAR modeling in predicting extractant performance, offering new mechanistic insights into Mn 2+ /Mg 2+ separation, and informing the development of novel extractants.