Different Strategies for Second Order Data Generation and Modeling. Applications to Solve Different Analytical Problems.

L. VERA-CANDIOTI; Y.S. CARO; M.M. DE ZAN; R. BRASCA; M.R. ALCARÁZ; M. MARCHISIO; F. PICCH; M.M. CÁMARA; A.V. SCHENONE; M.J. CULZONI; H.C. GOICOECHEA

Budapest

Congreso; XIII Chemometrics in Analytical Chemistry.; 2012

Second-order  data  enclose  the  so-called  "second-order  advantage",  which  allows predicting  the  concentration  of  the  analyte  of  interest  even  in  the  presence  of  unknown interferents, as well as enabling several analytes to be determined simultaneously [1]. In this report, results for several experimental data sets are presented in order to show the great  potentiality  of  the  second  order  data  modeled  with  convenient  algorithms  to  solve different  analytical  problems.  They  present  the  following  challenges  to  second-order algorithms: 1)  linear dependency due  to a kinetic  reaction  in one mode, 2) peak  shifts  (CE data),  and  3)  non-linearity.  In  all  of  these  cases,  deviations  from  the  ideal  trilinearity  are likely to occur due to changes in component profiles from sample to sample. Data  set  1  involves  five  fluoroquinolones  which  are  determined  in  environmental samples  (i.e.,  river  water)  by  using  capillary  electrophoresis  with  diode  array  detection. Multivariate  curve  resolution  with  alternating  least  squares  (MCR-ALS)  without  pre-treatment  outperformed  parallel  factor  analysis  (PARAFAC)  and  partial  least  squares followed by residual bilinearization (PLS/RBL) in profiles extraction and quantitation of the five analytes. Data set 2 includes fluorescence-time measurements made by creating a gradient within a flow  injection  system. These second order data were applied  to  resolve mixtures of two  antihistaminic  drugs  (loratadine  and  desloratadine)  in  serum  samples.  The  use  of  a surfactant was necessary  to obtain  the  selectivity  to differentiate both  spectra. This kind of data present  the problem of complete overlapping of profiles  in one data dimension, which can be regarded as a special and serious case of linear dependency [2]. The strategy involves the building of a MCR-ALS model composed of matrices augmented  in  the  temporal mode, i.e. spectra remain invariant while time profiles may change from sample to sample.  In Data set 3, fluorescence-time data were studied for the oxidation reaction of loratadine and desloratadine with potassium bromate, in order to determine both drugs in human serum samples. Linear dependency in both temporal and spectral modes precluded the use of MCR-ALS and PLS/RBL, providing better results than PARAFAC. Data set 4 consists of fluorescence-time data obtained for the oxidation reaction of three dyes with potassium bromate. The possibility of exploiting the second-order advantage from these  non-linear  second-order  data  could  be  reached  by  the  application  of  two  successive methods: the first one modeled the calibration and validation data removing the contribution of  unexpected  components,  and  the  second  one models  the  non-linear  relationship. MCR-ALS was the only strategy that retrieved reasonably accurate predictions.