Comparative study of different third-order data generation approaches

M.R. ALCARÁZ; M. MONTEMURRO; G.G. SIANO; M.J CULZONI; H.C. GOICOECHEA

Congreso; XVI Chemometrics in Analytical chemistry; 2016

Over the last years, it has been demonstrated that the increase of multiway data dimensions has a positive impact on analytical figures of merit, e.g. higher sensitivity, lower limits of detection and quantitation, better selectivity, among others. First- and second-order data analyses have become excellent tools for the resolution of complex samples which would result experimentally challenging from the univariatecalibration standpoint. On the other hand, even though no additional analytical advantages have been yet proved, third-order data analysis for analytical applications constitutes a field worth to be explored [1].Although multidimensional instrumental signals are easy to be obtained with the available modern instrumentation, and several chemometric algorithms have been successfully developed to solve multiway data problems, the way in which the multi-way data are generated may have a significant effect on the final results. In this work, a comparative study of different third-order data generation approaches was carried out. Three methods based on identical liquid chromatographic conditions but coupled to different emission and excitation fluorescence detection systems were developed for the quantitative analysis of antibiotics in aqueous matrices.The first approach included the collection of several fractions at the end of the chromatographic procedure by means of a custom-built device that allows to collect fractions in a 96-wells ELISA plate. Then, emission and excitation spectra were registered for every fraction by using a spectrofluorometer equipped with a plate reader accessory coupled to an optical fiber and a gated photomultiplier. In this way, 25 emission-excitation matrices (EEM, size: 17×25) were obtained for each chromatographic run. [2] A second strategy was developed by using a 10 μL flow-cell connected at the end of the chromatographic instrument and placed in a fast-scanning spectrofluorometer. Here, it was possible to register sequential EEMs for a unique chromatographic run. Since the fast-scannig spectrofluorometer takes only few seconds to register each EEM, it was necessary neither to stop the chromatographic flow nor to collect fractions after the chromatographic procedure, and 25 sequential EEMs (size: 7×45) were obtained for each chromatographic run. Finally, a multi-chromatographic run method involving a liquid chromatographic instrument coupled to a fast-scanning fluorescence detector which allowed to register time-emission fluorescence data matrices in a specific spectral range at a fixed excitation wavelength was developed. In order to build excitation-emission data matrices, eight chromatographic runs at different excitation wavelengths (time-emission matrix size: 45×150) were required for the same sample. The three methodologies aforementioned were evaluated using different algorithms, such as PARAFAC, APARAFAC, PLS-RTL and MCR-ALS, and selectivity, sensitivity, robustness, and time processing were evaluated. Since the data generation was different, each methodology required a particular data preprocessing including smoothing, peak alignment, and baseline correction, among others. Furthermore, due to differences in sensitivity provided by the implementation of a variety of detection mode it was necessary to assess several sample preparation methods in order to reach good analytical figures of merit.