Detection and Correction of Biased Results of Individual Analytes in Multicomponent Spectroscopic Analysis

Abstract

Simultaneous spectroscopic multicomponent analysis based on Beer's law requires the test sample to follow the hard model, independently of whether this model is built with the full spectrum or only with a few sensors selected according to an optimality criterion. We have developed a graphical method based on the net analytical signal concept to detect bias in the predicted results of individual analytes in test samples. When an interference, or other causes which produce bias, are detected, a moving window approach is used to select the subset of sensors that minimize the bias of the predicted results. The method has been validated with UV-vis spectra of binary clorophenol mixtures and of mixtures of four pesticides in water analyzed with a FIA system with diode array detection.

A new family of genetic algorithms for wavelength interval selection in multivariate analytical spectroscopy

“…Literature reported selection methods include simulated annealing [10], artificial neural networks [11], genetic algorithms (GAs) [12] and extensive searches using moving window strategies [8,13], among a great variety of other techniques (see Reference [14] and references cited therein). Comprehensive searches do not allow for the selection of multiple wavelength regions, because this activity is prohibitively time-consuming for spectra composed of a large number of data points.…”

Improved calculation of the net analyte signal in inverse multivariate calibration

“…The projection matrix can then be written using the transpose of Equation (5): a For PCR the matrices from the SVD algorithm (see Equation (16) in the Appendix) can be truncated for A latent variables to obtain U A , D A and P A . SIMPLS [24], NIPALS or other algorithms can be used for PLS regression.…”

“…c A is also the slope of the straight line with zero intercept that perfectly fits the points in a scatter plot of each element of r à A versus each element of s à . This plot is called the net analyte signal regression plot (NASRP) [16]. The slope can also be estimated from the least squares expression…”

“…Numerous papers have shown the importance of the multivariate NAS in the calculation of analytical figures of merit such as sensitivity [1][2][3][4][5][6][7][8][9][10][11][12][13], selectivity [1][2][3][4][5][6][7][8][9]11,12], signal-tonoise ratio [1][2][3]6,9,11] and limit of detection [1][2][3]5,9,14]. Applications of the NAS have already been described for detecting outliers [6,15,16], selecting variables [12,[15][16][17], representing multivariate models in two-dimensional plots ('pseudounivariate' presentations) [9][10][11]18] and interpreting prediction errors by means of the multivariate sensitivity [10,…”

“…This adds confidence to the values of the figures of merit calculated using the NAS. The modified equations also extend the use of the net analyte signal regression plot (described for classical models [16]) to PCR and PLS and offer an accurate pseudounivariate presentation of the model. This presentation can now be used both for interpreting and predicting.…”