One-class model with two decision thresholds for the rapid detection of cashew nuts adulteration by other nuts.

A green screening method to determine cashew nut adulteration with Brazilian nut, pecan nut, macadamia nut and peanut was proposed. The method was based on the development of a one-class soft independent modelling of class analogy (SIMCA) model for non-adulterated cashew nuts using near-infrared (NIR) spectra obtained with portable equipment. Once the model is established, the assignment of unknown samples depends on the threshold established for the authentic class, which is a key aspect in any screening approach. The authors propose innovatively to define two thresholds: lower model distance limit and upper model distance limit. Samples with distances below the lower threshold are assigned as non-adulterated with a 100% probability; samples with distance values greater than the upper threshold are assigned as adulterated with a 100% probability; and samples with distances within these two thresholds will be considered uncertain and should be submitted to a confirmatory analysis. Thus, the possibility of error in the sample assignment significantly decreases. In the present study, when just one threshold was defined, values greater than 95% for the optimized threshold were obtained for both selectivity and specificity. When two class thresholds were defined, the percentage of samples with uncertain assignment changes according to the adulterant considered, highlighting the case of peanuts, in which 0% of uncertain samples was obtained. Considering all adulterants, the number of samples that were submitted to a confirmatory analysis was quite low, 5 of 224 adulterated samples and 3 of 56 non-adulterated samples.

Multivariate qualitative methodology for semi-quantitative information. A case study: Adulteration of olive oil with sunflower oil.

This paper proposes a strategy to assess the performance of a multivariate screening method for semi-quantitative purposes. The adulteration of olive oil with sunflower oil was considered as a case study using fluorescence spectroscopy and two-class Partial Least Squares Discriminant Analysis (PLS-DA). Building the proper screening methodology based on two-class multivariate classification model involve setting the cut-off value for the adulterated class (class 2). So, four classification models were established for four levels of adulterant (cut-off). Model validation involved calculating the main quality parameters (sensitivity, specificity and efficiency) and three additional semi-quantitative parameters (limit of detection, detection capability and unreliability region). The probability of successfully recognizing non-adulterated samples as such was set by the main performance parameters of the two-class model. However, the probability of successfully recognizing adulterated samples as such was more accurately extracted from the performance characteristic curves (PCC) curves instead of just from the sensitivity of the adulterated class. The main performance parameters of the PLS-DA models increased as the cut-off level increased although after a particular value the increase was less pronounced. As an example, when the cut-off was changed from 5% to 20%, sensitivity changed from 70 to 93%, specificity changed from 87 to 97%, and efficiency changed from 78 to 95%. The same can be stated for the semi-quantitative parameter's decision limit and detection capability, which moved from 0 to 1.6 and from 17.7 to 21.6 (% of adulterant), respectively.

ROC curves for the optimization of one-class model parameters. A case study: Authenticating extra virgin olive oil from a Catalan protected designation of origin.

This paper proposes a ROC curve-based methodology to find optimal classification model parameters. ROC curves are implemented to set the optimal number of PCs to build a one-class SIMCA model and to set the threshold class value that optimizes both the sensitivity and specificity of the model. The authentication of the geographical origin of extra-virgin olive oils of Arbequina botanical variety is presented. The model was developed for samples from Les Garrigues, target class, Samples from Siurana were used as the non-target class. Samples were measured by FT-Raman with no pretreatment. PCA was used as exploratory technique. Spectra underwent pre-treatment and variables were selected based on their VIP score values. ROC curve and others already known criteria were applied to set the threshold class value. The results were better when the ROC curve was used, obtaining performance values higher than 82%, 75% and 77% for sensitivity, specificity and efficiency, respectively.

Authentication of the geographical origin of extra-virgin olive oil of the Arbequina cultivar by chromatographic fingerprinting and chemometrics.

This paper proposes to use chromatographic fingerprints coupled to multivariate techniques to authenticate the geographical origin of extra-virgin olive oils (EVOO) of the Arbequina botanical variety. This methodology uses the whole or part of the chromatogram as input data for the classification models but does not identify or quantify the chemicals constituents. Arbequina monovarietal EVOOs from three geographical origins were studied: two from adjacent European Protected Designation of Origin areas, Siurana and Les Garrigues, in Catalonia in the northeast of Spain; and the third from the south of Spain (Andalucia and Murcia). Three chromatographic fingerprints of each sample were obtained by both reverse and normal phase liquid chromatography coupled to charged aerosol detector (HPLC-CAD), and high temperature gas chromatography coupled to flame ionization detector [(HT)GC-FID]. Principal component analysis (PCA) was used as exploratory technique and soft independent modelling of class analogy (SIMCA) and partial least square-discriminant analysis (PLS-DA) were used as classification methods. High and low-level data fusion strategies were also applied to improve the classification results obtained when the data acquired from each analytical technique were separately used. The results were best for the PLS-DA model with low-level fusion of two techniques (HT)GC-FID with HPLC-CAD, independently of the phase mode. Sensitivity and specificity were 100% in almost all classes, error was 0% for all classes and an inconclusive ratio of just 4% was obtained for the Les Garrigues class due to double assignations.

Detection of adulterants in grape nectars by attenuated total reflectance Fourier-transform mid-infrared spectroscopy and multivariate classification strategies.

There is no any doubt about the importance of food fraud control, as it has implications in food safety and in consumer health. Focusing on fruit beverages, some types of adulterations have been detected more frequently, such as substitution with less expensive fruits. A methodology based on attenuated total reflectance Fourier-transform mid-infrared spectroscopy (ATR-FTIR) and multivariate classification was applied to detect whether grape nectars were adulterated by substitution with apple juice or cashew juice. A total of 126 samples were obtained and analyzed. Two strategies were proposed: one-class and multiclass approaches. Soft independent modeling of class analogy (SIMCA), partial least squares discriminant analysis (PLS-DA) and partial least squares density modeling (PLS-DM) were used to build the models. Among them, PLS-DA presented the best performance with a sensitivity and specificity of nearly 100%. The multiclass strategy was preferred if the adulterants to be studied are known because it provides additional information.

Variable selection for multivariate classification aiming to detect individual adulterants and their blends in grape nectars.

During the quality inspection control of fruit beverages, some types of adulterations can be detected, such as the addition or substitution with less expensive fruits. To determine whether grape nectars were adulterated by substitution with apple or cashew juice or by a mixture of both, a methodology based on attenuated total reflectance Fourier transform mid infrared spectroscopy (ATR-FTIR) and multivariate classification methods was proposed. Partial least squares discriminant analysis (PLS-DA) and soft independent modeling of class analogy (SIMCA) models were developed as multi-class methods (classes unadulterated, adulterated with cashew and adulterated with apple) with the full-spectra. PLS-DA presented better performance parameters than SIMCA in the classification of samples with just one adulterant, while poor results were achieved for samples with blends of two adulterants when using both classification methods. Three variable selection methods were tested in order to improve the effectiveness of the classification models: interval partial least squares (iPLS), variable importance in projection scores (VIP scores) and a genetic algorithm (GA). Variable selection methods improved the performance parameters for the SIMCA and PLS-DA methods when they were used to predict samples with only one adulterant. Only PLS-DA coupled with iPLS was able to classify samples with blends of two adulterants, providing sensitivity values between 100% and 83% at 100% specificity for the three studied classes.

HPLC-UV and HPLC-CAD chromatographic data fusion for the authentication of the geographical origin of palm oil.

Data fusion combined with a multivariate classification approach (partial least squares-discriminant analysis, PLS-DA) was applied to authenticate the geographical origin of palm oil. Data fusion takes advantage of the synergistic effect of information collected from more than one data source. In this study, data from liquid chromatography coupled to two detectors -ultraviolet (UV) and charged aerosol (CAD)- was fused by high- and mid-level data fusion strategies. Mid-level data fusion combines a few variables from each technique and then applies the classification technique. Principal component analysis and interval partial least squares were applied to obtain the variables selected. High-level data fusion combines the PLS-DA classification results obtained individually from the chromatographic technique with each detector. Fuzzy aggregation connective operators were used to make the combinations. Prediction rates varied between 73% and 98% for the individual techniques and between 87% and 100% and 93% and 100% for the mid- and high-level data fusion strategies, respectively.

Determining performance parameters in qualitative multivariate methods using probability of detection (POD) curves. Case study: Two common milk adulterants.

A strategy for determining performance parameters of two-class multivariate qualitative methods was proposed. As case study, multivariate classification methods based on mid-infrared (MIR) spectroscopy coupled with the soft independent modelling of class analogy (SIMCA) technique for detection of hydrogen peroxide and formaldehyde in milk were developed. From the outputs (positive/negative/inconclusive) of the samples, which were unadulterated and adulterated at target value, the main performance parameters were obtained. Sensitivity and specificity values for the unadulterated and adulterated classes were satisfactory. Inconclusive ratios 12% and 21%, respectively, for hydrogen peroxide and formaldehyde were obtained. To evaluate the performance parameters related to concentration, Probability of Detection (POD) curves were established, estimating the decision limit, the capacity of detection and the unreliability region. When inconclusive outputs were obtained, two additional concentration limits were defined: the decision limit with inconclusive outputs and the detection capability with inconclusive outputs. The POD curves showed that for concentrations below 3.7gL-1 of hydrogen peroxide and close to zero of formaldehyde, the chance of giving a positive output (adulterated sample) was lower than 5%. For concentrations at or above 11.3gL-1 of hydrogen peroxide and 10mgL-1 of formaldehyde, the probability of giving a negative output was also lower than 5%.

Detection of several common adulterants in raw milk by MID-infrared spectroscopy and one-class and multi-class multivariate strategies.

A sequential strategy was proposed to detect adulterants in milk using a mid-infrared spectroscopy and soft independent modelling of class analogy technique. Models were set with low target levels of adulterations including formaldehyde (0.074g.L-1), hydrogen peroxide (21.0g.L-1), bicarbonate (4.0g.L-1), carbonate (4.0g.L-1), chloride (5.0g.L-1), citrate (6.5g.L-1), hydroxide (4.0g.L-1), hypochlorite (0.2g.L-1), starch (5.0g.L-1), sucrose (5.4g.L-1) and water (150g.L-1). In the first step, a one-class model was developed with unadulterated samples, providing 93.1% sensitivity. Four poorly assigned adulterants were discarded for the following step (multi-class modelling). Then, in the second step, a multi-class model, which considered unadulterated and formaldehyde-, hydrogen peroxide-, citrate-, hydroxide- and starch-adulterated samples was implemented, providing 82% correct classifications, 17% inconclusive classifications and 1% misclassifications. The proposed strategy was considered efficient as a screening approach since it would reduce the number of samples subjected to confirmatory analysis, time, costs and errors.

FT-Raman and NIR spectroscopy data fusion strategy for multivariate qualitative analysis of food fraud.

Two data fusion strategies (high- and mid-level) combined with a multivariate classification approach (Soft Independent Modelling of Class Analogy, SIMCA) have been applied to take advantage of the synergistic effect of the information obtained from two spectroscopic techniques: FT-Raman and NIR. Mid-level data fusion consists of merging some of the previous selected variables from the spectra obtained from each spectroscopic technique and then applying the classification technique. High-level data fusion combines the SIMCA classification results obtained individually from each spectroscopic technique. Of the possible ways to make the necessary combinations, we decided to use fuzzy aggregation connective operators. As a case study, we considered the possible adulteration of hazelnut paste with almond. Using the two-class SIMCA approach, class 1 consisted of unadulterated hazelnut samples and class 2 of samples adulterated with almond. Models performance was also studied with samples adulterated with chickpea. The results show that data fusion is an effective strategy since the performance parameters are better than the individual ones: sensitivity and specificity values between 75% and 100% for the individual techniques and between 96-100% and 88-100% for the mid- and high-level data fusion strategies, respectively.

A tutorial on the validation of qualitative methods: from the univariate to the multivariate approach.

This tutorial provides an overview of the validation of qualitative analytical methods, with particular focus on their main performance parameters, for both univariate and multivariate methods. We discuss specific parameters (sensitivity, specificity, false positive and false negative rates), global parameters (efficiency, Youden's index and likelihood ratio) and those parameters that have a quantitative connotation since they are usually associated to concentration values (decision limit, detection capability and unreliability region). Some methodologies that can be used to estimate these parameters are also described: the use of contingency tables for the specific and global parameters and the performance characteristic curve (PCC) for the ones with quantitative connotation. To date, PCC has been less commonly used in multivariate methods. To illustrate the proposals summarized in this tutorial, two cases study are discussed at the end, one for a univariate qualitative analysis and the other for multivariate one.

Partial least squares density modeling (PLS-DM) - a new class-modeling strategy applied to the authentication of olives in brine by near-infrared spectroscopy.

A new class-modeling method, referred to as partial least squares density modeling (PLS-DM), is presented. The method is based on partial least squares (PLS), using a distance-based sample density measurement as the response variable. Potential function probability density is subsequently calculated on PLS scores and used, jointly with residual Q statistics, to develop efficient class models. The influence of adjustable model parameters on the resulting performances has been critically studied by means of cross-validation and application of the Pareto optimality criterion. The method has been applied to verify the authenticity of olives in brine from cultivar Taggiasca, based on near-infrared (NIR) spectra recorded on homogenized solid samples. Two independent test sets were used for model validation. The final optimal model was characterized by high efficiency and equilibrate balance between sensitivity and specificity values, if compared with those obtained by application of well-established class-modeling methods, such as soft independent modeling of class analogy (SIMCA) and unequal dispersed classes (UNEQ).

Validation of multivariate screening methodology. Case study: detection of food fraud.

Multivariate screening methods are increasingly being implemented but there is no worldwide harmonized criterion for their validation. This study contributes to establish protocols for validating these methodologies. We propose the following strategy: (1) Establish the multivariate classification model and use receiver operating characteristic (ROC) curves to optimize the significance level (α) for setting the model's boundaries. (2) Evaluate the performance parameter from the contingency table results and performance characteristic curves (PCC curves). The adulteration of hazelnut paste with almond paste and chickpea flour has been used as a case study. Samples were analyzed by infrared (IR) spectroscopy and the multivariate classification technique used was soft independent modeling of class analogies (SIMCA). The ROC study showed that the optimal α value for setting the SIMCA boundaries was 0.03 in both cases. The sensitivity value was 93%, specificity 100% for almond and 98% for chickpea, and efficiency 97% for almond and 93% for chickpea.

Multivariate screening in food adulteration: untargeted versus targeted modelling.

Two multivariate screening strategies (untargeted and targeted modelling) have been developed to compare their ability to detect food fraud. As a case study, possible adulteration of hazelnut paste is considered. Two different adulterants were studied, almond paste and chickpea flour. The models were developed from near-infrared (NIR) data coupled with soft independent modelling of class analogy (SIMCA) as a classification technique. Regarding the untargeted strategy, only unadulterated samples were modelled, obtaining 96.3% of correct classification. The prediction of adulterated samples gave errors between 5.5% and 2%. Regarding targeted modelling, two classes were modelled: Class 1 (unadulterated samples) and Class 2 (almond adulterated samples). Samples adulterated with chickpea were predicted to prove its ability to deal with non-modelled adulterants. The results show that samples adulterated with almond were mainly classified in their own class (90.9%) and samples with chickpea were classified in Class 2 (67.3%) or not in any class (30.9%), but no one only as unadulterated.

Figures of merit of a SERS method for Sudan I determination at traces levels.

A substrate for Surface-Enhanced Raman Scattering spectroscopy (SERS), electropolished Al, is proposed as a tool for a rapid and low cost determination of Sudan I. This dye has been used as an additive in some foodstuffs but it is now banned because of the health risk associated with its carcinogenic and mutagenic properties. Despite the presence of fluorescence, Raman spectra of Sudan I can be obtained using excitation lasers at 633 and 785 nm. To get rid of the spectral noise and fluorescence background, Savitzky-Golay smoothing and polynomial corrections were applied, respectively. The Raman signal was proved to be enhanced. A linear dependence was found between the logarithmic intensity at 1598 cm(-1) peak versus the logarithmic concentration. The figures of merit were studied obtaining high sensitivity and low detection limits (10(-7) M). A multivariate exploratory analysis (PCA) was used to study the ability of SERS to distinguish Sudan I from other similar compounds. Therefore, results show that SERS is a potential tool to determine Sudan I quickly and effectively.

Quantitative analysis of the effect of zidovudine, efavirenz, and ritonavir on insulin aggregation by multivariate curve resolution alternating least squares of infrared spectra.

Quantification of the effect of antiretroviral drugs on the insulin aggregation process is an important area of research due to the serious metabolic diseases observed in AIDS patients after prolonged treatment with these drugs. In this work, multivariate curve resolution alternating least squares (MCR-ALS) was applied to infrared monitoring of the insulin aggregation process in the presence of three antiretroviral drugs to quantify their effect. To evidence concentration dependence in this process, mixtures at two different insulin:drug molar ratios were used. The interaction between insulin and each drug was analysed by (1)H NMR spectroscopy. In all cases, the aggregation process was monitored during 45 min by infrared spectroscopy. The aggregates were further characterised by scanning electron microscopy (SEM). MCR-ALS provided the spectral and concentration profiles of the different insulin-drug conformations that are involved in the process. Their feasible band boundaries were calculated using the MCR-BANDS methodology. The kinetic profiles describe the aggregation pathway and the spectral profiles characterise the conformations involved. The retrieved results show that each of the three drugs modifies insulin conformation in a different way, promoting the formation of aggregates. Ritonavir shows the strongest promotion of aggregation, followed by efavirenz and zidovudine. In the studied concentration range, concentration dependence was only observed for zidovudine, with shorter aggregation time obtained as the amount of zidovudine increased. This factor also affected the aggregation pathway.

Standardization of UV-visible data in a food adulteration classification problem.

This study evaluates the performance of multivariate calibration transfer methods in a classification context. The spectral variation caused by some experimental conditions can worsen the performance of the initial multivariate classification model but this situation can be solved by implementing standardization methods such as Piecewise Direct Standardization (PDS). This study looks at the adulteration of culinary spices with banned dyes such as Sudan I, II, III and IV. The samples are characterised by their UV-visible spectra and Partial Least Squares-Discriminant Analysis (PLS-DA) is used to discriminate between unadulterated samples and samples adulterated with any of the four Sudan dyes. Two different datasets that need to be standardised are presented. The standardization process yields positive classification results comparable to those obtained from the initial PLS-DA model, in which high classification performance was achieved.

Surface Enhanced Raman Spectroscopy (SERS) and multivariate analysis as a screening tool for detecting Sudan I dye in culinary spices.

Raman spectroscopy combined with multivariate analysis was evaluated as a tool for detecting Sudan I dye in culinary spices. Three Raman modalities were studied: normal Raman, FT-Raman and SERS. The results show that SERS is the most appropriate modality capable of providing a proper Raman signal when a complex matrix is analyzed. To get rid of the spectral noise and background, Savitzky-Golay smoothing with polynomial baseline correction and wavelet transform were applied. Finally, to check whether unadulterated samples can be differentiated from samples adulterated with Sudan I dye, an exploratory analysis such as principal component analysis (PCA) was applied to raw data and data processed with the two mentioned strategies. The results obtained by PCA show that Raman spectra need to be properly treated if useful information is to be obtained and both spectra treatments are appropriate for processing the Raman signal. The proposed methodology shows that SERS combined with appropriate spectra treatment can be used as a practical screening tool to distinguish samples suspicious to be adulterated with Sudan I dye.

1H NMR variable selection approaches for classification. A case study: the determination of adulterated foodstuffs.

Whenever dealing with large amount of data as is the case of a NMR spectrum, carrying out a variable selection before applying a multivariate technique is necessary. This work applies various variable selection techniques to extract relevant information from (1)H NMR spectral data. Three approaches have been chosen, because each is based on very different foundations. The first method, called Xdiff, is based on calculating the normalized differences between the mean spectrum of a class considered to be the reference and the spectra of each sample. The second approach is the interval Partial Least Squares method (iPLS), which investigates the influential zones of the spectra that contains the most discriminating predictors calculating local PLS-DA models on narrow intervals. The last one is Genetic Algorithms (GAs) which finds the optimal variables from a random initial subset of variables by means of an iterative process. The performance of each variable selection strategy is determined by the classification results obtained when multiclass Partial Least Squares-Discriminant Analysis is applied. This study has been applied to NMR spectra of culinary spices that might be adulterated with banned dyes such as Sudan dyes (I-IV). The three techniques give neither the same number nor the same selected variables, but they do select a common zone from the spectra containing the most discriminating variables. All three techniques give satisfactory classification and prediction results, being higher than 95% with iPLS and GA and around 89% with Xdiff, therefore the three variable selection techniques are suitable to be used with NMR data in the determination of food adulteration with Sudan dyes as well as the specific type of adulterant used (I-IV).

1H NMR and UV-visible data fusion for determining Sudan dyes in culinary spices.

Two data fusion strategies (variable and decision level) combined with a multivariate classification approach (Partial Least Squares-Discriminant Analysis, PLS-DA) have been applied to get benefits from the synergistic effect of the information obtained from two spectroscopic techniques: UV-visible and (1)H NMR. Variable level data fusion consists of merging the spectra obtained from each spectroscopic technique in what is called "meta-spectrum" and then applying the classification technique. Decision level data fusion combines the results of individually applying the classification technique in each spectroscopic technique. Among the possible ways of combinations, we have used the fuzzy aggregation connective operators. This procedure has been applied to determine banned dyes (Sudan III and IV) in culinary spices. The results show that data fusion is an effective strategy since the classification results are better than the individual ones: between 80 and 100% for the individual techniques and between 97 and 100% with the two fusion strategies.