Abstract

The basis of interpretive optimisation in liquid chromatography is the prediction of resolution, from appropriate solute retention models. The reliability of the process depends critically on the quality of the experimental design. This work develops, validates and applies a general methodology aimed to evaluate the quality of any training experimental design, which will be applied in Part II to design optimisation. The methodology is based on the systematic evaluation of the uncertainties associated to the prediction of retention times in comprehensive scans of both isocratic and gradient experimental conditions. It is able to evaluate comprehensively experimental designs of arbitrary complexity. Five common training experimental designs were used to model the retention, according to the Linear Solvent Strength (LSS) and the Neue-Kuss (NK) equations, using a set of 14 sulphonamides of different polarity. The results are presented in terms of relative uncertainties in predictions, which provide significant and interpretable results. The magnitude of such uncertainties, together with the systematic, coherent and logical changes observed at increasing solute hydrophobicity, give support to the results. The NK model gave smaller errors and unbiased predictions, whereas the LSS model gave rise to lack of fit. Isocratic training designs, which are widely accepted as the most informative, are confirmed as the best. As a general conclusion, gradients are predicted with intrinsically smaller uncertainties, independently of the training experimental design. In addition, gradients are more insensitive than isocratic predictions with regard to the type of training design used. Isocratic predictions deteriorate quickly with mobile phase composition. This explains the better performance of gradient predictions, even with biased models.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call