A new approach based on inversion of a partial least squares model searching for a preset analytical target profile....

The paper shows a procedure for selecting the control method parameters (factors) to obtain a preset 'analytical target profile' when a liquid chromatographic technique is going to be carried out for the simultaneous determination of five bisphenols (bisphenol-A, bisphenol-S, bisphenol-F, bisphenol-Z and bisphenol-AF), some of them regulated by the European Union. The procedure has three steps. The first consists of building a D-optimal combined design (mixture-process design) for the control method parameters, which are the composition of the ternary mobile phase and its flow rate. The second step is to fit a PLS2 model to predict six analytical responses (namely, the resolution between each pair of consecutive peaks, and the initial and final chromatographic time) as a function of the control method parameters. The third final step is the inversion of the PLS2 model to obtain the conditions needed for attaining a preset analytical target profile. The computational inversion of the PLS2 prediction model looking for the Pareto front of these six responses provides a set of experimental conditions to conduct the chromatographic determination, specifically 22% of water, mixed with 58% methanol and 20% of acetonitrile, keeping the flow rate at 0.66 mL min^-1. These conditions give a chromatogram with retention times of 2.180, 2.452, 2.764, 3.249 and 3.775 min for BPS, BPF, BPA, BPAF and BPZ, respectively, and excellent resolution among all the chromatographic peaks. Finally, the analytical method is validated under the selected experimental conditions, in terms of trueness and precision. In addition, the detection capability for the five bisphenols were: 596, 334, 424, 458 and 1156 μg L^-1, with probabilities of false positive and of false negative equal to 0.05.