Abstract
Introduction: Predictability of partial least squares regression (PLSR) method is improved by four multivariate signal processing modelling approaches; including genetic algorithm PLS (GAPLS), net analyte processing PLS (NAP-PLS), orthogonal signal correction PLS (OSC-PLS) and direct orthogonal signal correction PLS (DOSC-PLS). The objective of the introduced work is to establish a comparison among proposed chemometric models; indicating the constituent algorithm of each and setting a comparison of analysis results.
Method: The proposed models are compared through stability indicating analysis of mixtures of metopimazine (MTP) and its reported degradation products via handling spectrofluorimetric data, at excitation wavelength of 268 nm and working emission range of 420-520 nm. The degradation products include the oxidative degradation product (OMTP) and the alkaline hydrolysis degradation product (AMTP).
A 3-factor 5-level experimental design was set; ending up with a training set of 25 mixtures containing different ratios of the interfering components. A test set composed of 12 mixtures was implemented to validate the predictive ability of the proposed models. Leave one out- cross validation procedure (LOO-CV) was used to predict optimum number of PLS components.
Result: The 4 introduced models were successfully applied to assay MTP in raw material and the best results were given by GA-PLSR (test set 99.00% ± 2.980).
Conclusion: The 4 models were applied to analysis of pharmaceutical tablets, then statistically compared to a reported spectrofluorimetric method; showing no significant difference regarding accuracy and precision; indicating the ability of the suggested models to be trusted for routine quality control analysis of pharmaceutical product.
Keywords: GA-PLS, NAP-PLS, OSC-PLS, DOSC-PLS, metopimazine, spectrofluorimetry.
Graphical Abstract
30
1