Current Analytical Chemistry - Volume 8, Issue 2, 2012

Full text available.

An investigation was carried out to assess the suitability of an approach based on a metric generalization of the Partial Least Square Regression to identify and discriminate chain dependence relationships between geology, soil properties and grape composition. The study area was Telesina Valley (Campania Region, southern Italy), a foremost vineyard district with the Falanghina grape cultivar. Five geopedological units, with different assemblages of geological substrates and pedotypes, were surveyed. Soil surface horizons were sampled from 85 sites, representing the widest variability of geological and pedological features. Our results allowed, for each geopedological unit, to categorise the specific variables influencing grape quality, and to quantify their individual contribution. In particular, geopedological units with volcanic (Andisols) and more or less developed (Inceptisols) soils on recent (Olocenic) colluvial debris, as well as with Andisols and organic matter-rich soils (Mollisols) on older Pleisto-Olocenic pyroclastites, express the best performance of the main components of wine taste and flavour, as tartaric acid, malic acid and titratable acidity of grape. Specifically, these last were significantly dependent on soil features as fine texture, neutral-alkaline pH, and appreciable content of soil organic matter and high CEC.

In a recent study of about 6500 white and red Portuguese wines, a formal method of classification was considered based on 11 key chemical and physical properties. These properties were all deemed as being important for influencing the quality of wine that is produced in the area. In this paper, we explore these physiochemical characteristics and begin by providing a glimpse into understanding the link between these characteristics and wine quality. A more formal analytical approach is considered that involves undertaking a multiple-regression analysis of these characteristics. A key aspect of this paper is to provide a graphical comparison of this formal method of classification with the classification given to the wines by three experienced assessors. The graphical approach considered in this paper is correspondence analysis applied to the confusion matrices formed by cross-classifying the wine levels given by the assessors and those specified by more formal classification methods. The use of two-dimensional plots and three-dimensional plots obtained from the correspondence analysis of these matrices is explored, and the advantages of other graphical means, including the dendrogram, to visualise the association between these categories are briefly considered.

Evaluating the accuracy of classification methods for rating wines based on their physical and chemical characteristics is an important part of the wine industry. An ability to accurately predict the quality of wine based on this information provides vast opportunity for applications in other areas of science. Recently, a method was proposed that utilises data mining techniques for the prediction of the quality of wines based solely on their physicochemical properties and compared these results with those classifications obtained by experienced assessors. In this paper we explore an analytical approach to evaluating the accuracy of these classification methods using new advances in the area of statistical modelling.

In this paper we present a four-step procedure for performing a Tucker2 analysis on three-way data in the case a number of observations are missing. The procedure consists of (1) creating multiple complete data sets via multiple imputation of the missing data, (2) analysing these data sets with the Tucker2 model, (3) combining both the row and the column component matrices of these analyses to create centroid solutions using Generalised Procrustes analyses, and (4) using these centroid solutions to find the associated core slices appertaining to the centroid solutions. This procedure will produce both the basic parameters of the Tucker2 model and estimates of their variability due to the missing data. Chromatographic data are used as an illustration.

Routinely, the multi-response Partial Least-Squares (PLS) is used in regression and classification problems showing good performances in many applied studies. In this paper, we aim to present PLS via spline functions focusing on supervised classification studies and showing how PLS methods historically belong to L2 boosting family. The theory of the PLS boost models is presented and used in classification studies. As a natural enrichment of linear PLS boost, we present its multi-response non-linear version by univariate and bivariate spline functions to transform the predictors. Three case studies of different complexities concerning soils and its products will be discussed, showing the gain in diagnostic provided by the non-linear additive PLS boost discriminant analysis compared to the linear one.

Bagging and boosting have become increasingly important ensemble methods for combining models in the data mining and machine learning literature. We review the basic ideas of these methods, propose a new robust boosting algorithm based on a non-convex loss function and compare the performance of these methods to both simulated and real data sets with and without contamination.

The analysis of high dimensional dataset is recurrently used in chemometrics where the data are presented in the form of digitized spectra (NIR). Statistical tool, as Discriminant Analysis, is frequently used in this field to classify object in predefined categories. But, by the fact that this kind of dataset presents the number of statistic units relatively small in comparison to the number of variables, the classical Discriminant Analysis can not be applied. In this paper, the authors, present a strategy to choose an optimal subset of predictors to perform Discriminant Analysis on NIR data in partial least squares framework.

In chemistry, multiblock datasets are easily encountered with variables of different natures, or measured at different times for example, here, we use the sequential multiblock regression method GOMCIA-PLS1 to predict quantitative variables from several predictors gathered according to their nature and used simultaneously. In this article, it will be applied to predict a chemical variable from Near Infrared Spectrometry (NIRS) chemical and thermolyze data measured on different tobacco samples. The multiblock GOMCIA-PLS1 method is compared to other methods and its good performances are shown.

This study demonstrated the use of visible-near infrared (vis-NIR) reflectance spectroscopy and partial least squares regression (PLSR) for the effective analysis of important properties of Mediterranean soils from southern Italy. Understanding soil properties is an essential pre-requisite for sustainable land management. Assessment of these properties has long been gained through conventional laboratory analysis, which is considered costly and time consuming. Therefore, there is a need to develop alternative cheaper and faster techniques for soil analysis. In recent years, special attention has been given to vis-NIR reflectance spectroscopy and chemometrics. In this study we evaluated the potential of vis-NIR spectroscopy and PLSR for prediction of chemical and physical properties [sand, silt and clay, organic carbon (OC), total nitrogen (N), cation exchange capacity (CEC), and calcium carbonate (CaCO3)] of soils representative of three Mediterranean agro-ecosystems from the Campania region, southern Italy. We performed the analysis for each agroecosystem separately (local predictions) and for the combined ones (regional prediction). PLSR is one of the most popular modelling techniques used in chemometrics and is commonly used for quantitative spectroscopic analysis. We derived PLSR models, which were validated using an independent subset of data that was not used in the modelling. The accuracy of the calibrations and validations for the different soil properties were assessed using the root mean squared error (RMSE) and the relative percent deviation (RPD). Our results showed that regional and local predictions are from very good to excellent for OC (RPD of validation = 2.36 ÷ 3.03) and clay content (RPD = 2.31 ÷ 2.95). For the remaining properties, RPD values ranged from 1.40 ÷ 2.07 (poor/fair-very good), for regional predictions, to 1.10 ÷ 2.33 (poor-very good), for local predictions.

Polycyclic Aromatic Hydrocarbons (PAHs) are a group of carcinogenic contaminants widespread in the environment. PAHs are produced by both anthropogenic and natural processes. Difficulties exist in identifying their origins. This paper reports a practical application of Principal Component Analysis (PCA) and Principal Component Regression (PCR) to identify the pyrolytic, petrogenic and diagenesis sources of PAH pollution in the Sarno River and Estuary. Nicknamed “the most polluted river in Europe”, the Sarno River originates in south-western Italy and has a watershed of about 715 km2. PCA indicated that the PAH contamination in the Sarno River and Estuary resulted from a mixed pattern. The first principal component (PC1) had significant positive loading in high molecular weight PAHs. This profile of PAH usually includes products of high temperature combustion/pyrolitic processes, reflecting the effects of traffic pyrolysis. The second principal component (PC2) had significant positive loading in two-to-four ring PAHs. So, PC2 may be considered as components from petrogenic sources. PC3 was characterized by a high loading of perylene, thought to originate from diagenetic alteration of perylenequinone pigment or some other organic matter. Therefore, this factor can be considered as natural-origin PAHs. In the PCR, the regression coefficients for components 1-3 were 66.6, 40.4 and 19.5, respectively. In this application, the PCR was a very useful statistical technique for handling the problem of multicollinearity. Results from the application of PCR have been compared with Partial Least Square (PLS) and no significant differences were reported in the prediction errors and latent variables available by PCR and PLS.

A continuum approach for canonical correlation analysis based on a single tuning parameter is proposed. One of its noteworthy features is that it is straightforward and simple. It establishes a bridge between unfold Principal Components Analysis and Canonical Correlation Analysis. Interesting properties related to the method of analysis are also discussed. This makes it possible to highlight the rationale behind the continuum approach by showing that it aims at realizing a compromise between achieving the investigation of the relationships between two datasets and setting up a stable model. The strategy of analysis is exemplified on the basis of a case study aiming at linking metabolomics and proteomics data, on the one hand, and discriminating groups of individuals using metabolomics data, on the other hand.

Due to the well established difference in the pharmaco/toxicological profile of many amino acid enantiomers, and also for typifying the food quality and origin, the exact knowledge of their presence and relative ratio in foodstuffs, is a matter of growing interest. In this setting, with an interest in identifying the presence of D-amino acids in a selected set of cheese samples, and with the aim to introduce a fast and easily accessible chromatographic procedure, we analyzed six cheese extracts with a CLEC-based chiral stationary phase (CLEC-CSP). The CLEC analyses were run without any preor post-column derivatization of the amino acidic mixture. The successful chemo- and enantioseparation were contemporarily achieved with the use of a dynamically coated CSP (C-CSP) based on the S-trityl-L-cysteine (L-STC) as the chiral selector. With the applied CLEC procedure, the presence of D-Ala, D-Asp and D-Glu was diagnosed in all the analyzed samples and then confirmed via conventional chiral gas chromatographic (CGC) analysis. A certain degree of peak overlapping was found to be the main drawback of the simplified sample analysis, which is nevertheless balanced by the advantages of the rapid detection.

This paper proposes a hybrid chemometric method named DOSC-WPT-PLS, which is based on partial least squares (PLS) regression combined with direct orthogonal signal correction (DOSC) and wavelet packet transform (WPT) as pre-processing tools for the simultaneous spectrophotometric kinetic determination of Mn (II), Ag (I) and Fe (III). This method combined with kinetic approach is used for the first time and no past reference has been found for this determination. The improved PLS regression algorithm was developed by adding a preprocessor based on DOSC and WPT to enhance the prediction ability. The kinetic chemometric approach that combines kinetic-catalytic method with DOSC-WPTPLS does not require a detailed kinetic model of the chemical system to obtain the order of reaction and rate constants. The relative standard errors of prediction (RSEP) obtained for all components using DOSC-WPT-PLS, WPTPLS and PLS were compared for evaluating their predictive capability. Results obtained from the DOSC-WPT-PLS method were significantly better, and proved that it outperformed the WPTPLS and PLS methods.