Prediction of the retention factor in cetyltrimethylammonium bromide modified micellar electrokinetic chromatography using a machine learning approach

Abstract

Capillary electrophoresis (CE) is an analytical technique widely applied in clinical, industrial, and scientific laboratories. Discussion of scientists' and specialists' concerns regarding the superiority of CE and more frequently used high-performance liquid chromatography (HPLC) is well known. With several advantages like a short analysis time, high efficiency, and low reagents consumption (mostly organic solvent), CE is considered a “green” alternative to HPLC. The relationship between retention and molecule structure has paid attention practically from the very beginning of separation methods. It can be established using a quantitative structure retention relationship (QSRR) method. The main goal of our investigation is to fill the gap related to QSRR analyses for the cetrimonium bromide (CTAB) micellar electrokinetic chromatography (MEKC) system using a heterogeneous set of 89 model molecules. The genetic algorithm (GA) supported the selection of theoretical descriptors that quantitatively describe target solutes. Comparison of linear and non-linear algorithms has been performed. Finally, QSRR models using partial least squares regression (PLS) and support vector regression (SVR) have been developed, and their performances were evaluated. The obtained results clearly indicate that machine learning (ML) models can be used as supportive tools in predicting retention in the CTAB-MEKC system. Among investigated models, the best performance showed GA-PLS. This finding has been proved by leave-one-out cross-validation and external validation procedures. Notably, the established models also give a view into the molecular mechanism of interaction between molecules and CTAB-formed micelles. Our investigations confirmed that descriptors related to the lipophilicity of molecules are the most significant factors in these types of interactions.