Multiple-column RP-HPLC retention modelling based on solvatochromic or theoretical solute descriptors

Recently, we have proposed an approach to multi-column RP-HPLC retention modelling under isocratic conditions based on a combination of five molecular descriptors, the volume fraction of organic modifier in the mobile phase and a column descriptor, simultaneously considered as dependent variables of artificial neural network (ANN) regression. The column descriptor, in particular, was identified with the observed average retention of the solutes used in calibration extrapolated to pure water as the mobile phase. The ANN-based model was seen to accurately describe retention on a pool of octadecylsiloxane-bonded (C(18)) columns in a wide range of mobile phase composition. Reliability of this approach is further examined here by analysing the retention data of Reta et al. (Anal. Chem. 1999, 71, 3484-3496) referring to 17 aromatic compounds collected in water-methanol mobile phases at the compositions 45, 50, 55 and 60%v/v of methanol with eight different columns based on various hydrocarbon, fluorocarbon and aromatic bonded stationary phases. Further, in this study we compare the explanatory capability of two different kinds of molecular descriptors: the well-known solvatochromic descriptors and theoretical descriptors extracted by genetic algorithm variable selection from the large set provided by the popular software Dragon.