Comparison of correlation ranking and eigenvalue ranking unfolded principal component regression for direct determination of naproxen in human serum using excitation–emission matrix fluorescence spectroscopy

Abstract

Using principal components as input of a multivariate calibration model always brings up the question what subset of PCs gives the best reliable regression model with the most predictability. In this current study, two different methods are presented and discussed for the determination of naproxen (NAP) in serum based on principal component regression. At the first step, the fluorescence landscapes of NAP in serum with excitation wavelengths from 235 to 293 nm and emission wavelengths in the range 320–420 nm were obtained. The resulted excitation emission fluorescence matrices were unfolded and subjected to principal component analysis, and a linear regression model was built to model the relationship between the extracted principal components (PCs) and the concentrations. Two approaches were applied to select the most relevant principal components: eigenvalue ranking and correlation ranking. The results of this study showed that the importance of each PC to the multivariate calibration is not correctly described by its eigenvalue, but by its influence on the prediction of the dependent variable. Since it is possible for a PC to have a low variance in original dataset (low eigenvalue) but a high correlation with the dependent variable, eigenvalue ranking selection should not be employed. The proposed method is an interesting alternative to the traditional techniques normally used for determining naproxen such as HPLC. Compared to HPLC method, the proposed EV-PCR and CR-PCR methods were rapid, easy and of low cost for the quantification of NAP in serum using simple UV–Vis spectrofluorimetry.