Influence of descriptor database selection on modeling retention factors in capillary micellar and microemulsion electrokinetic chromatography using the solvation parameter model

Abstract

Abraham's solvation parameter model has been widely used to model retention in capillary micellar and microemulsion electrokinetic chromatography systems. To fit or predict retention factors in separation systems experimentally determined compound descriptors are required. These are typically obtained from two sources: the Abraham and Wayne State University (WSU) compound descriptor databases. Alternatively, compound descriptors can be estimated from molecular structures using computational, machine learning, and group contribution approaches without having to resort to experimental methods for compounds without assigned values in either database. Here, we compare descriptor quality for the Abraham and WSU experimental compound descriptor databases as well as computational approaches using group contribution and machine learning approaches for modelling retention in capillary micellar and microemulsion electrokinetic chromatography systems. It is shown that the WSU compound descriptor database affords a more accurate description for the retention of varied compounds while descriptors based on a machine learning approach more accurately modeled retention factors than a group contribution approach. Model standard errors and coefficients of determination for WSU descriptors varied from 0.046 to 0.116 and 0.976 to 0.996, respectively, while Abraham descriptors varied from 0.048 to 0.166 and 0.953 to 0.995, respectively. Model standard errors and coefficients of determination for machine learning descriptors ranged from 0.086 to 0.116 and 0.979 to 0.981, respectively, whereas group contribution descriptors ranged from 0.090 to 0.181 and 0.942 to 0.979, respectively. Descriptor quality is important when modeling retention in capillary micellar and microemulsion electrokinetic chromatography systems.