Prediction of gradient retention from the linear solvent strength (LSS) model, quantitative structure‐retention relationships (QSRR), and artificial neural networks (ANN)

Abstract

AbstractThe linear solvent strength (LSS) model combined with quantitative structure‐retention relationships (QSRR) and artificial neural network (ANN) analysis has been shown to permit approximate prediction of the gradient high‐performance liquid chromatography (HPLC) retention time for any analyte on a once‐characterized column. The approach applies well to the reversed‐phase HPLC mode with a methanol‐water (buffer) eluent of linearly changing composition. Its suitability was tested for a representative series of structurally diverse analytes. In this approach the determination of retention times, t R, in two gradient runs for a predesigned model series of 15 analytes is first needed. Next, model QSRR equations describing t R in terms of analyte structure are derived to characterize the HPLC systems of interest. To quantitatively characterize the structure of the analytes the following three structural descriptors from molecular modeling are employed: total dipole moment; electron excess charge of the most negatively charged atom; and water‐accessible molecular surface area. Using these data a general QSRR equation is derived which is valid for a given column/eluent system. Next, having the structural descriptors for any analyte to be chromatographed in such a characterized HPLC system, one employs the previously derived general QSRR equation to calculate the analyte's retention time. The expected gradient retention time for any gradient conditions can be calculated by means of appropriate LSS equations. Independent of the standard QSRR calculation procedure based on multiple regression analysis (MRA), predictions of gradient retention times were performed by means of artificial neural networks (ANN). It has been found that the predictive power of ANN is similar to that of MRA. The combined LSS/QSRR approach has been demonstrated to provide approximate, yet otherwise unattainable, a priori predictions of gradient retention of analytes based solely on their chemical formulae. That way a rational chemometric basis for a systematic optimization of chromatographic separations has been elaborated as an alternative to the trial‐and‐error method normally applied at present.