M.S. Sánchez

Capability of detection of an analytical method evaluating false positive and false negative (ISO 11843) with partial least squares

Analytical techniques based on soft multivariate calibrations (as those which provide first and second order analytical signals necessarily are) remain outside the field of application of the ISO norms related to capability of detection. In this work, a complete solution for the problem of applying ISO norm 11843 to soft calibration (for instance, one or multi-way partial least squares (PLS)) is provided. The methodological procedure is applied to different case studies which implies different analytical techniques.

Handling intrinsic non-linearity in near-infrared reflectance spectroscopy

Abstract The relationship between absorption in the near-infrared (NIR) spectral region and the target analytical parameter is frequently of the non-linear type. The origin of the non-linearity can be widely varied and difficult to identify. In some cases, the relationship between absorption and the analytical parameter of interest is intrinsically non-linear owing to the very chemical nature of the sample or analyte concerned. In this work, various multivariate calibration procedures were tested with a view to overcoming intrinsic non-linearity in NIR reflectance. An approach to solving the problem is suggested. Calibration was done, after transformation of spectra, by using linear and non-linear techniques. The linear calibration techniques used are partial least squares (PLS) regression (with and without variable selection), linear PLS with X projection (LP-PLS) and stepwise polynomial principal component (SWP-PC) regression. Non-linear calibration methods included polynomial PLS (PPLS) and artificial neural networks (ANNs). Results were compared on the basis of NIR spectra for ampicillin trihydrate samples, where the simultaneous presence of crystallization water and surface moisture gives rise to intrinsic non-linearity that affects the determination of the total water content in the sample. The best results were obtained by using the non-linear calibration techniques.

Efficiency of multi-layered feed-forward neural networks on classification in relation to linear discriminant analysis, quadratic discriminant analysis and regularized discriminant analysis

Abstract The efficiency of multi-layered feed-forward networks (MLF) on classification is evaluated by applying them to simulated data. The classes are normal multivariate with three different structures for the matrix of covariance. For each of them a complete factorial design, 2 3 , was performed, with a replicated central point in order to study the effect of the relationships objects—variables, noise—signal and distance between centroids. The results were compared to those obtained by applying linear discriminant analysis, quadratic discriminant analysis and regularized discriminant analysis to the same sets of data. The comparison was carried out by an ANOVA of the experimental designs and by principal components and correspondence analysis.

Applicability of high-absorbance MIR spectroscopy in industrial quality control of reformed gasolines

Abstract Partial least squares (PLS), polynomial partial least squares (polynomial-PLS), locally weighted regression (LWR) and genetic inside neural network (GINN) algorithms were used to develop models for predicting motor octane number (MON) from non-leaded and catalytically reformed gasolines. Medium infrared (mid-infrared) spectra were obtained on liquid samples and chemometrically processed in order to get acceptable predictive models which allow their use for routine industrial quality monitoring. As MIR spectra currently present peaks with high absorbances, the presence and influence of nonlinearities was sought comparing the broadly-used PLS method with several other algorithms specially designed to cope with such influences (polynomial-PLS, local regression and neural networks). Their prediction abilities; i.e., stability and global prediction error when predicting new samples as well as their usefulness for routine industrial control were studied.

Performance of multi-layer feedforward and radial base function neural networks in classification and modelling

Abstract Neural networks have been used in multiple applications, but as a kind of black box for dealing with problems where there is no a priori information about the data. This means that the model is constructed based solely upon information obtained from the data themselves. This seems to be a good property but makes it difficult to validate the models obtained. The classification properties of neural classifiers are usually described by the percentage of correctly classified objects in a test set. Since these straight methods are only based on discrimination, no information can be obtained in a statistical way. In this paper, on a simulated data set, two different types of neural networks, MLF (multi layer feedforward) and RBF (radial base function), are applied to solve a classification problem. The modelling ability, stability and reproducibility of this kind of networks are studied based on various different networks independently trained on the same data set with a predetermined value for the sensibility and specificity. Robustness to different kinds of error is also studied by means of Monte Carlo simulations adding noise at different levels and from different theoretical distributions. Further to this, an analysis based on principal components is carried out to study the apparently different networks obtained. The simulation studies reveal that both types of networks perform well enough to reproduce the input space. For RBF networks, due to the local approach, the study showed some properties related to sensibility and specificity which are relevant in practical problems.

Quality control decisions with near infrared data

Abstract In this paper, as an alternative to multivariate regression methods, quality control tasks are posed as a decision problem: a sample is acceptable (this means that it follows its way to market) or not (then, it should be carefully examined according to laboratory procedures). The parameter to control is the content of water in samples of ampicillin trihydrate, based on near-infrared (NIR) spectra obtained from reflectance measurements. For modelling purposes, Genetic Inside Neural Network (GINN) is used. GINN is a neural network-based tool designed to perform the best possible decision by means of simultaneous optimisation of both type-I and type-II errors. Further, this training is made without imposing any condition on the distribution of data (nonparametric) and under nonlinear conditions.

Pareto-optimal front as a tool to study the behaviour of experimental factors in multi-response analytical procedures.

This work presents a methodology to analyse the behaviour of an analytical procedure, above all when optimization of the procedure is needed. The methodology starts by the design of an experiment suitable to fit response surfaces to some analytical responses of interest in the problem being studied. Then, a pareto-optimal front is estimated that accounts for the optimal possibly trading-off solutions among the responses. The analysis of the behaviour of the optimal values of the response surfaces and the experimental conditions that provide these values allows going deeply into the analytical procedure.

Psychophysical parameters of colour and the chemometric characterisation of wines of the certified denomination of origin ‘Rioja’

Abstract Colour is one of the most important characteristics of a wine. To measure it, the International Organisation for Wine (OIV) proposes the use of the so-called CieLab parameters: a ∗ , red/green chromaticity; b ∗ , yellow/blue chromaticity; and L ∗ , clarity. However, the need for including the psychophysical parameters: C ∗ , chroma, H ∗ , tone, and S ∗ , saturation has been suggested by some opinions. The six parameters are internationally normalised and they are obtained from the absorption spectrum in the visible range. The aim of this work is to show the interest of the second option through the results of multivariate classification and modelling analysis. The models built with only the three CieLab variables showed to be insufficient for a good characterisation of the colour of the young red wines. When the additional variables C ∗ , H ∗ and S ∗ are introduced, the specificity of the UNEQ models for both categories are improved, from 48 to 70% and from 75 to 81%, respectively. The European Community (EC) has defined claret and rose wines, allowing the consumption of the first only in Spain. In addition, the EC has stated as fraudulent the practice of blending white and red wines to be sold as rose wines. Using wines from the denomination of origin Rioja, the SIMCA models built with the all six variables rejected the blended wines while those built with only three variables did not.

Improving the visualization of the Pareto-optimal front for the multi-response optimization of chromatographic determinations

The paper shows tools to visualize and more easily interpret the effect that some experimental factors may exert on analytical responses of interest when optimization of several responses is needed. It is based on an adaptation of the parallel coordinate plot, a tool for graphical representation of points in multidimensional spaces that, theoretically and contrary to the usual Cartesian plots, does not have limits in the dimension of the points being depicted. The joint use of the Pareto-optimal solutions and their visualization allows a deeper knowledge about the problem at hand as well as the wise selection of the conditions of experimental factors for achieving specific goals about the responses. Although the methodology is for a general use, the procedure, its interpretation and usefulness is shown with several analytical cases in chromatography. The first one refers to the experimental conditions to obtain simultaneously the maximum allowable area for both the peak of the malachite green and its metabolite leucomalachite green in fish by liquid chromatography with tandem mass spectrometry detection (LC-MS/MS). The second one is about the simultaneous determination of steroid hormones estrone and 17-α-ethinylestradiol by gas chromatography-mass spectrometry (GC/MS). In the last case, the chromatographic separation by GC/MS of the diastereoisomers, α- and β-estradiol is needed taking into account that these hormones have the same mass fragments.

On the construction of experimental designs for a given task by jointly optimizing several quality criteria: Pareto-optimal experimental designs.

Experimental designs for a given task should be selected on the base of the problem being solved and of some criteria that measure their quality. There are several such criteria because there are several aspects to be taken into account when making a choice. The most used criteria are probably the so-called alphabetical optimality criteria (for example, the A-, E-, and D-criteria related to the joint estimation of the coefficients, or the I- and G-criteria related to the prediction variance). Selecting a proper design to solve a problem implies finding a balance among these several criteria that measure the performance of the design in different aspects. Technically this is a problem of multi-criteria optimization, which can be tackled from different views. The approach presented here addresses the problem in its real vector nature, so that ad hoc experimental designs are generated with an algorithm based on evolutionary algorithms to find the Pareto-optimal front. There is not theoretical limit to the number of criteria that can be studied and, contrary to other approaches, no just one experimental design is computed but a set of experimental designs all of them with the property of being Pareto-optimal in the criteria needed by the user. Besides, the use of an evolutionary algorithm makes it possible to search in both continuous and discrete domains and avoid the need of having a set of candidate points, usual in exchange algorithms.