Direct orthogonal signal correction as data pretreatment in the classification of clinical lots of creams from near infrared spectroscopy data.

Direct orthogonal signal correction (DOSC) is applied to correct for major variance sources such as temperature effects, time influences and instrumental differences in near infrared (NIR) data. The samples analysed are creams containing different concentrations of an active drug. The final aim is to classify the samples according to their concentration of active compound. Having performed DOSC on the data, it is not necessary anymore to apply sophisticated chemometric techniques to correct for temperature or time effects and to attribute the samples to their respective concentration classes. Moreover, the application of DOSC on the NIR spectra recorded on two different instruments shows that this method can be considered as a valuable alternative for the standardisation in classification applications. Since the applied algorithm tends to overfit, in a second part of this paper, a comparison is made with an algorithm designed by Westerhuis, which should overcome this problem. Although the calibration set results show that the overfitting has been partially corrected for by the latter algorithm, the test set results did not improve significantly.

Near-infrared spectroscopy applications in pharmaceutical analysis.

Near-infrared (NIR) spectroscopy is a fast and non-destructive analytical technique that offers many advantages for a broad range of industrial applications. In this work, we reviewed recent developments in the pharmaceutical domain where it can be applied from raw material identification to final product release. The characteristics of NIR allow the technique to be implemented as a process analytical technology (PAT). Moreover, recent instrumental developments open the perspectives of numerous applications in the NIR imaging area. After "Introduction", according to their subject, the applications are discussed in the parts "Identification", "Water content", "Assay" and "Other applications".

Chemometric analysis of soil pollution data using the Tucker N-way method.

N-way methods, particularly the Tucker method, are often the methods of choice when analyzing data sets arranged in three- (or higher) way arrays, which is the case for most environmental data sets. In the future, applying N-way methods will become an increasingly popular way to uncover hidden information in complex data sets. The reason for this is that classical two-way approaches such as principal component analysis are not as good at revealing the complex relationships present in data sets. This study describes in detail the application of a chemometric N-way approach, namely the Tucker method, in order to evaluate the level of pollution in soil from a contaminated site. The analyzed soil data set was five-way in nature. The samples were collected at different depths (way 1) from two locations (way 2) and the levels of thirteen metals (way 3) were analyzed using a four-step-sequential extraction procedure (way 4), allowing detailed information to be obtained about the bioavailability and activity of the different binding forms of the metals. Furthermore, the measurements were performed under two conditions (way 5), inert and non-inert. The preferred Tucker model of definite complexity showed that there was no significant difference in measurements analyzed under inert or non-inert conditions. It also allowed two depth horizons, characterized by different accumulation pathways, to be distinguished, and it allowed the relationships between chemical elements and their biological activities and mobilities in the soil to be described in detail.

Class identity assignment for amphetamines using neural networks and GC-FTIR data.

An exploratory analysis was performed in order to evaluate the feasibility of building of neural network (NN) systems automating the identification of amphetamines necessary in the investigation of drugs of abuse for epidemiological, clinical and forensic purposes. A first neural network system was built to distinguish between amphetamines and nonamphetamines. A second, more refined system, aimed to the recognition of amphetamines according to their toxicological activity (stimulant amphetamines, hallucinogenic amphetamines, nonamphetamines). Both systems proved that discrimination between amphetamines and nonamphetamines, as well as between stimulants, hallucinogens and nonamphetamines is possible (83.44% and 85.71% correct classification rate, respectively). The spectroscopic interpretation of the 40 most important input variables (GC-FTIR absorption intensities) shows that the modeling power of an input variable seems to be correlated with the stability and not with the intensity of the spectral interaction. Thus, discarding variables only because they correspond to spectral windows with weak absorptions does not seem be not advisable.

Chemometric treatment of vanillin fingerprint chromatograms. Effect of different signal alignments on principal component analysis plots.

This study describes the chemometric treatment of vanillin fingerprint chromatograms to distinguish vanillin from different sources. Prior to principal component analysis, which is used to discriminate vanillin from different origins, the fingerprints are aligned. Three alignment algorithms are tested, correlation optimized warping (COW), target peak alignment (TPA) and semi-parametric time warping (STW). The performance of the three algorithms is evaluated and the effect of the different alignments on the PCA score plots is investigated. The alignment obtained with STW differs somewhat from that with COW and TPA. However, equivalent score plots were obtained regarding the different vanillin groups.

Evaluation of chemometric techniques to select orthogonal chromatographic systems.

Several chemometric techniques were compared for their performance to determine the orthogonality and similarity between chromatographic systems. Pearson's correlation coefficient (r) based color maps earlier were used to indicate selectivity differences between systems. These maps, in which the systems were ranked according to decreasing or increasing dissimilarities observed in the weighted-average-linkage dendrogram, were now applied as reference method. A number of chemometric techniques were evaluated as potential alternative (visualization) methods for the same purpose. They include hierarchical clustering techniques (single, complete, unweighted-average-linkage, centroid and Ward's method), the Kennard and Stone algorithm, auto-associative multivariate regression trees (AAMRT), and the generalized pairwise correlation method (GPCM) with McNemar's statistical test. After all, the reference method remained our preferred technique to select orthogonal and identify similar systems.

Local factor analysis of rank-deficient reaction systems.

The analysis of spectral measurement data sets using local factor analysis (LFA) requires the rank of the sub-matrix under study to be equal to the number of absorbing species present in the associated sub-system. However, because of mass balance or kinetic constraints, LFA will fail if local rank deficiency occurs. A local rank deficiency sub-system may be present in a global full-rank reaction system or a rank-deficient one. In this paper, the problems occurring when using window target-testing factor analysis (WTTFA), one type of the LFA methods, in a local rank-deficient situation are shown. A new augmented WTTFA (AWTTFA) is then proposed for the correct use of WTTFA when rank deficiency occurs. Principles of this new method have been demonstrated by a simulated kinetic system and an industrial batch data set.

Automatic program for peak detection and deconvolution of multi-overlapped chromatographic signals part I: peak detection.

A series of two papers describing a procedure for automated peak deconvolution is presented. The goal is to develop a package of routines that can be used by non-experienced users. Part I (this paper) concerns peak detection, whereas Part II is dedicated to the deconvolution itself. In this first part, the most interesting features of the peak detection algorithms, which precede the deconvolution step, are outlined. High-order derivatives provide valuable information to assess the number of underlying compounds under a given peak cluster. A smoothing technique was found essential to compute properly the derivatives, since the noise is amplified when differences are calculated. The Savitsky-Golay smoother was applied in combination with the Durbin-Watson criterion to automate the window size selection. This strategy removed the noise without loosing valuable information. In some cases, it was found preferable to split the chromatogram in different elution regions, and apply the Durbin-Watson test and the Savitsky-Golay smoother to each region, separately. The derivatives allowed obtaining estimates of both peak parameters and the corresponding ranges for each eluting compound to be used in the deconvolution. An algorithm oriented to compare peaks from different chromatograms is also presented to perform deconvolution, using information from several related chromatograms.

Automatic program for peak detection and deconvolution of multi-overlapped chromatographic signals part II: peak model and deconvolution algorithms.

Several interlinked algorithms for peak deconvolution by non-linear regression are presented. These procedures, together with the peak detection methods outlined in Part I, have allowed the implementation of an automatic method able to process multi-overlapped signals, requiring little user interaction. A criterion based on the evaluation of the multivariate selectivity of the chromatographic signal is used to auto-select the most efficient deconvolution procedure for each chromatographic situation. In this way, non-optimal local solutions are avoided in cases of high overlap, and short computation times are obtained in situations of high resolution. A new algorithm, fitting both the original signal and the second derivatives is proved to avoid local optima in intermediate coelution situations. This allows achieving the global optimum without the need of background knowledge by the user. A previously reported peak model, a Gaussian with a polynomial standard deviation whose complexity can be modulated to enhance the fitting quality, was applied. However, the original formulation was modified to account baseline outside the peak region. Also, the optimal model complexity was auto-selected via error propagation theory. The method is able to process simultaneously several related chromatograms. The software was tested with both simulated and experimental chromatograms obtained with monolithic silica columns.

Prediction of total green tea antioxidant capacity from chromatograms by multivariate modeling.

In this paper, a fast strategy for determining the total antioxidant capacity of Chinese green tea extracts is developed. This strategy includes the use of experimental techniques, such as fast high-performance liquid chromatography (HPLC) on monolithic columns and a spectrophotometric approach to determine the total antioxidant capacity of the extracts. To extract the chemically relevant information from the obtained data, chemometrical approaches are used. Among them there are correlation optimized warping (COW) to align the chromatograms, robust principal component analysis (robust PCA) to detect outliers, and partial least squares (PLS) and uninformative variable elimination partial least squares (UVE-PLS) to construct a reliable multivariate regression model to predict the total antioxidant capacity from the fast chromatograms.

Development of a ginkgo biloba fingerprint chromatogram with UV and evaporative light scattering detection and optimization of the evaporative light scattering detector operating conditions.

A fingerprint chromatogram of a standardized Ginkgo biloba extract is developed on a monolithic silica column using a ternary gradient containing water, iso-propanol and tetrahydrofuran. For the detection, UV and evaporative light scattering (ELS) detectors are used, the latter allowing detection of the poor UV absorbing compounds as ginkgolides (A-C and J) and bilobalide in the extract. The complementary information between the UV and ELS fingerprint is evaluated. The ELS detector used in this study can operate in an impactor 'on' or 'off' mode. For each mode, the operating conditions such as the nebulizing gas flow rate, the drift tube temperature and the gain are optimized by use of three-level screening designs to obtain the best signal-to-noise (S/N) ratio in the final ELS fingerprint chromatogram. In both impactor modes, very similar S/N ratios are obtained for the nominal levels of the design. However, optimization of the operating conditions resulted, for both impactor modes, in a significant increase in S/N ratios compared to the initial evaluated conditions, obtained from the detector software.

Prediction of gastro-intestinal absorption using multivariate adaptive regression splines.

Multivariate adaptive regression splines (MARS) and a derived method two-step MARS (TMARS) were used for modelling the gastro-intestinal absorption of 140 drug-like molecules. The published absorption values for these molecules were used as response variable and calculated molecular descriptors as potential explanatory variables. Both methods were compared and their potential use in quantitative structure-activity relationship (QSAR) context evaluated. The predictive abilities of the models were studied using different sequences of Monte Carlo cross validation (MCCV). It was shown that both types of models had good predictive abilities and that for the data used, MARS gave better results than TMARS. It could be concluded that both methods could be valuable for QSAR modelling.

Classification of drugs in absorption classes using the classification and regression trees (CART) methodology.

Classification and regression trees (CART) were evaluated for their potential use in a quantitative structure-activity relationship (QSAR) context. Models were build using the published absorption values for 141 drug-like molecules as response variable and over 1400 molecular descriptors as potential explanatory variables. Both the role of two- and three-dimensional descriptors and their relative importance were evaluated. For the used dataset, CART models showed high descriptive and predictive abilities. The predictive abilities were evaluated based on both cross-validation and an external test set. Application of the variable ranking method to the models showed high importances for the n-octanol/water partition coefficient (logP) and polar surface area (PSA). This shows that CART is capable of selecting the most important descriptors, as known from the literature, for the absorption process in the intestinal tract.

Orthogonality and similarity within silica-based reversed-phased chromatographic systems.

The starting point of this study was a current set of 32 chromatographic systems used to select initial conditions for method development to determine the impurity profile of a drug. The system exhibiting the best selectivity is then selected for further method development. In this current set eight silica-based phases are applied in conjunction with four mobile phases at different pH. In order to save time and resources, the possibilities for a meaningful subset selection were investigated. The most differing systems in terms of selectivity, in other words only the most orthogonal systems, need to be selected. Since the stationary phases are all silica-based, the selectivity differences are examined within a more homogeneous group than if, for instance, also zirconia- or polymer-based columns would be involved. To select the subset of systems also the best overall separation performances are taken into account. The selection is based both on the HPLC-DAD data of a generic set of 68 drugs, and on the LC-MS-DAD results for a mixture of 15 drugs, less different in structure. The orthogonality is evaluated using weighted-average-linkage dendrograms and color maps, both created from the Pearson-correlation coefficients r between normalized retention times r. The Derringer's desirability functions are applied to define the systems with the best overall separation performances. Proposals for different representative subsets of the initial 32 systems are made.

Chemometrical exploration of the wet precipitation chemistry from the Austrian Monitoring Network (1988-1999).

The present paper deals with the application of different chemometric methods to an environmental data set derived from the monitoring of wet precipitation in Austria (1988-1999). These methods are: principal component analysis (PCA); projection pursuit (PP); density-based spatial clustering of application with noise (DBSCAN); ordering points to identify the clustering structures (OPTICS); self-organizing maps (SOM), also called the Kohonen network; and the neural gas (NG) network. The aim of the study is to introduce some new approaches into environmetrics and to compare their usefulness with already existing techniques for the classification and interpretation of environmental data. The density-based approaches give information about the occurrence of natural clusters in the studied data set, which, however, do not occur in the case presented here; information about high-density zones (very similar samples) and extreme samples is also obtained. The partitioning techniques (clustering, but also neural gas and Kohonen networks) offer an opportunity to classify the objects of interest into several defined groups, the patterns of ionic concentration of which can be studied in detail. The visual aids, such as the color map and the Kohonen map, for each site are very helpful in understanding the relationships between samples and between samples and variables. All methods, and in particular projection pursuit, give information about samples with extreme characteristics.

Boosting partial least squares.

A difficulty when applying partial least squares (PLS) in multivariate calibration is that overfitting may occur. This study proposes a novel approach by combining PLS and boosting. The latter is said to be resistant to overfitting. The proposed method, called boosting PLS (BPLS), combines a set of shrunken PLS models, each with only one PLS component. The method is iterative: the models are constructed on the basis of the residuals of the responses that are not explained by previous models. Unlike classical PLS, BPLS does not need to select an adequate number of PLS components to be included in the model. On the other hand, two parameters must be determined: the shrinkage value and the iteration number. Criteria are proposed for these two purposes. BPLS was applied to seven real data sets, and the results demonstrate that it is more resistant than classical PLS to overfitting without loosing accuracy.

Updating multivariate calibrations with the delaunay triangulation method.

Multivariate calibrations must be updated when new samples show different spectral characteristics. In this paper, we discuss how to do this when the calibration is performed with a topological multivariate calibration method based on Delaunay triangulation (DT). The updating leads either to the expansion of the original calibration set or to the creation of a new local model. Outliers in the new samples with respect to the original calibration set are first detected and divided in two groups, namely, marginal outliers, which are considered to be extensions of the calibration set and are used for updating the calibration set, and true outliers. If a sufficient number of true outliers are found to be situated close enough to each other, they can form the basis for a new local model. Several updating simulations performed on a real data set show that the updating procedure performs well. The results for prediction with the DT method after updating are comparable to or better than those after updating with partial least squares (PLS) and it is concluded that, in many cases, the DT method is a valuable alternative for multivariate calibration.

Improving QSAR models for the biological activity of HIV Reverse Transcriptase inhibitors: Aspects of outlier detection and uninformative variable elimination.

The goal of this study is to derive a methodology for modeling the biological activity of non-nucleoside HIV Reverse Transcriptase (RT) inhibitors. The difficulties that were encountered during the modeling attempts are discussed, together with their origin and solutions. With the selected multivariate techniques: robust principal component analysis, partial least squares, robust partial least squares and uninformative variable elimination partial least squares, it is possible to explore and to model the contaminated data satisfactory. It is shown that these techniques are versatile and valuable tools in modeling and exploring biochemical data.

Fast monolithic micellar liquid chromatography: an alternative drug permeability assessing method for high-throughput screening.

Several methods estimating the partitioning over biological membranes and thus the biological activity of potential oral drug molecules have been developed and are described in the literature. A previous study suggested that fast micellar liquid chromatography on a monolithic column could be one of them. For a set of diverse pharmaceuticals, retention by this fast chromatographic method was determined, besides other parameters also thought or established to describe oral permeability or absorption, e.g., from the Caco-2 permeability method. In view of a high-throughput determination of membrane permeability, a study was made of which information fast micellar liquid chromatography is providing and to what degree this system can replace other methods, i.e., deliver similar information. The retention with this fast method, which is mainly based on hydrophobic interactions, proved useful to sort substances into classes of Caco-2 and percent intestinal absorption.

Multivariate adaptive regression splines (MARS) in chromatographic quantitative structure-retention relationship studies.

The multivariate adaptive regression splines (MARS) methodology was applied to build quantitative structure-retention relationships (QSRRs). The response (dependent variable) in the MARS models consisted of the logarithms of the extrapolated retention factors (log k(w)) of 83 structurally diverse drugs on a Unisphere PBD column, using isocratic elutions at pH 11.7. A set of 266 molecular descriptors was used as predictor (independent) variables in the MARS model building. The optimal MARS model uses 34 basis functions to describe the retention and has acceptable predictive properties for new objects. The molecular descriptors included in the model describe hydrophobicity, molecular size, complexity, shape and polarisability. Some additional MARS models were created using alternative strategies. These include models with log P as the single predictor and models obtained with only the three most important molecular descriptors. The use of classification and regression trees (CART) as feature selection technique for predictor variables used in the MARS model was also investigated. Further, it is also studied whether allowing quadratic terms instead of interaction terms might lead to better MARS models.