P. Vankeerberghen

Comparison of semirobust and robust partial least squares procedures

Abstract Three different robust partial least squares (PLS) alternatives are compared in order to reduce the effect of outlying data. Although iteratively reweighted least squares, based on the biweight mean, offers efficiency in a variety of situations, in higher dimensions, there are still considerable deviations from the ordinary PLS solution without outliers.

The application of Kohonen neural networks to diagnose calibration problems in atomic absorption spectrometry

In atomic absorption spectrometric measurements calibration lines are measured daily. These lines are not always acceptable. They can, for instance, contain outliers, have a bad precision or can be curved. To evaluate the quality of those lines a method which gives a fast diagnosis is recommended. In this study the use of Kohonen neural networks was examined as an automated procedure to classify these calibration lines. The results were compared with those obtained using a decision support system which uses classical statistical methods to classify the lines. The prediction capabilities of both approaches relative to a visual inspection and classification was found to be comparable, or even slightly better for the Kohonen networks, depending on the training set used. For both techniques a prediction error rate of <10% was obtained, relative to a visual classification.

The quality coefficient as a tool in decisions about the quality of calibration in graphite furnace atomic absorption spectrometry

Abstract Vankeerberghen, P. and Smeyers-Verbeke, J., 1992. The quality coefficient as a tool in decisions about the quality of calibration in graphite furnace atomic absorption spectrometry. Chemometrics and Intelligent Laboratory Systems , 15:195–202. A quality coefficient (QC) is proposed as a probe of the quality of calibrations for use in decision support systems in graphite furnace atomic absorption spectrometry (GF-AAS). The quality coefficients described in the literature assume that the relative standard deviation over the calibration range is constant. From the variance functions established for copper and manganese it follows that, in the straight part of the GF-AAS calibration line, a constant standard deviation seems more justified than a constant coefficient of variation. A QC based on a constant standard deviation is therefore proposed. This was found to give a better characterization of the quality of straight calibration lines in GF-AAS.

Robust regression and outlier detection for non-linear models using genetic algorithms

Abstract Experimental data such as calibration and pharmacokinetic data can be contaminated with outliers. Robust regression based on the calculation of the least median of squared residuals (LMS) is robust to the presence of outliers and is used for outlier detection. The original LMS program only handles models which are linear in the parameters. A genetic algorithm can be used to obtain the LMS parameters for models that are non-linear in the parameters. In this work the feasibility of using genetic algorithms for LMS is demonstrated by means of curved analytical calibration and pharmacokinetic data contaminated with outliers.

Some robust statistical procedures applied to the analysis of chemical data

Abstract Vankeerberghen, P., Vandenbosch, C., Smeyers-Verbeke, J. and Massart, D.L., 1991. Some robust statistical procedures applied to the analysis of chemical data. Chemometrics and Intelligent Laboratory Systems , 12: 3–13. Real data are often subject to outliers, deviations from normality, etc., that violate assumptions made using classical tests or make the usual statistical descriptive parameters difficult to use. Different approaches to obtain robust statistical description or testing are described. They concern robust estimators of central location and dispersion, hypothesis tests using ranking and randomization methods and robust regression based on ranking.

Robust orthogonal regression for the outlier detection when comparing two series of measurement results

Abstract The application of a robust orthogonal regression procedure is described for the outlier detection during the evaluation of a method comparison study and, in general, when comparing two series of measurement results. The procedure is based on the least median of the squared orthogonal residuals (orthogonal LMS). It is concluded that the orthogonal LMS procedure is a useful screening tool to detect outlying values in a data set. Especially for data sets with extreme values it is recommended to combine the robust data evaluation with a visual display. For the comparison of heteroscedastic methods, the program can easily be adapted to a weighted orthogonal LMS variant.

Decision support system for run suitability checking and explorative method validation in electrothermal atomic absorption spectrometry

A decision support system for the run suitability check and an explorative method validation for electrothermal atomic absorption calibration lines is described. The first sub-system, the run suitability check, investigates whether the quality of calibration and standard addition lines meets the pre-established standards. When this is not the case, the system identifies the problem for which remedies may be proposed, when appropriate. The second sub-system, the explorative method validation, is based on parts of the run suitability check and investigates whether candidate methods are promising or need further exploration.

Modelling of solute retention in the reversed-phase high-performance liquid chromatography system with the chemically bonded 3-cyanopropyl stationary phase

We investigated the applicability of three models of solute retention (devised by Snyder et al., Schoenmakers et al., and Kowalska) for reversed-phase high-performance liquid chromatography (RP-HPLC) for description of the chromatographic behaviour of 25 test solutes in the RP-HPLC systems with the 3-cyanopropyl stationary phase and the methanol+water eluent. The employed test solutes considerably differed with respect to molecular structure, as they belonged to chemical classes of (i) benzene and alkylbenzenes, (ii) hydroxyaromatics, and (iii) crown ethers. In the case of the chemically bonded stationary phases, the 3-cyanopropyl phase included, one encounters the mixed mechanism of solute retention with the contributory effects of partition and adsorption. From our investigations, it comes out that the retention model proposed by Schoenmakers et al. (i.e., model II) is the best suited for description of the chromatographic behaviour of solutes on such packings, while the remaining two models (i.e., models I and III) perform less accurately. An additional investigation was performed with aid of the relationship between the retention parameters (i.e., log k) of the employed test solutes and the logarithms of their partition coefficients (log P) in the octanol/water system (as calculated from the hydrophobic fragmental constants after Rekker), depending on the quantitative composition of the binary eluent. The greater is the deviation of the log k vs. log P relationship from linearity (as monitored by the decreasing values of the respective linear correlation coefficients, r), the less pronounced becomes the contribution of partition and hence the importance of adsorption for the overall retention increases. Thus, the log k vs. log P study was performed in order to follow the partition↔adsorption balance with the changing volume proportions of methanol in the mixed eluent, and ultimately to link an extent of contamination of the nominal reversed-phase partition mechanism by adsorption with the predictive power of models I–III. Finally, an effort was undertaken aimed at demonstration of a considerably enhanced performance of model III, when applied to the experimental conditions that corresponds better with the theoretical assumptions of this approach.

Description and implementation of the system for the run suitability check and explorative method validation

Abstract This paper has two goals: the first is to describe the external representation to the user of the decision support system for the run suitability check and explorative method validation in atomic absorption spectrometry. This is done by means of a few representative examples. Once the visual representation of the system is illustrated, the second part of this paper describes the internal architecture of the system. The paper illustrates the evolution from the first to the current implementation as a Toolbook-dynamic link library hybrid system. This work demonstrates that these different development tools can be combined in such a way that the benefits of both tools are obtained. Since the current system is a re-write and further development of an existing program, the enhancements are briefly outlined.