Quantitative structure/eluent–retention relationships in reversed-phase high-performance liquid chromatography based on the solvatochromic method

Abstract

Some predictive approaches aimed at modelling the combined effect of solute molecular structure and mobile phase composition on retention in reversed-phase high-performance chromatography (RP-HPLC) have been developed in the literature. These models are established for a given binary eluent (normally acetonitrile–water or methanol–water) by non-linear (curvilinear or artificial neural network) regression assuming as the mobile phase descriptor the volume fraction φ of the organic modifier. In the present investigation, we propose a model applicable simultaneously to acetonitrile–water and methanol–water eluents. To this end, the Kamlet-Taft solvatochromic descriptors of the eluent and the solvatochromic descriptors of the analytes are considered as the input variables of a multi-layer artificial neural network (ANN) providing the solute retention as the response. This approach is applied to a set of 31 molecules analyzed with five different columns in the φ range 20–70 % at 10 % steps for both acetonitrile- and methanol-containing mobile phases. For each column, an ANN-based model is built using retention data of 25 molecules selected by the Kennard-Stones algorithm while retention data of the unselected six solutes are considered in the final evaluation of predictive performance of the trained network. To test cross-eluent prediction, the network optimized for a given column was successively trained with data collected in eight out of 12 eluents and applied to deduce retention in the four remaining mobile phases. The results reveal that RP-HPLC behavior of external solutes is quite accurately modelled in the whole explored composition range of acetonitrile– and methanol–water mobile phases. Moreover, the model exhibits a promising capability of deducing retention of external solutes even in unknown eluents.