Sensitivity and specificity of PLS-class modelling for five sensory characteristics of dry-cured ham using visible and near infrared spectroscopy

Machine learning methods in near infrared spectroscopy for predicting sensory traits in sweetpotatoes

It has been established that near infrared (NIR) spectroscopy has the potential of estimating sensory traits given the direct spectral responses that these properties have in the NIR region. In sweetpotato, sensory and texture traits are key for improving acceptability of the crop for food security and nutrition. Studies have statistically modelled the levels of NIR spectroscopy sensory characteristics using partial least squares (PLS) regression methods. To improve prediction accuracy, there are many advanced techniques, which could enhance modelling of fresh (wet and un-processed) samples or nonlinear dependence relationships. Performance of different quantitative prediction models for sensory traits developed using different machine learning methods were compared. Overall, results show that linear methods; linear support vector machine (L-SVM), principal component regression (PCR) and PLS exhibited higher mean R2 values than other statistical methods. For all the 27 sensory traits, calibration models using L-SVM and PCR has slightly higher overall R2 (x¯ = 0.33) compared to PLS (x¯ = 0.32) and radial-based SVM (NL-SVM; x¯= 0.30). The levels of orange color intensity were the best predicted by all the calibration models (R2 = 0.87 - 0.89). The elastic net linear regression (ENR) and tree-based methods; extreme gradient boost (XGBoost) and random forest (RF) performed worse than would be expected but could possibly be improved with increased sample size. Lower average R2 values were observed for calibration models of ENR (x¯ = 0.26), XGBoost (x¯ = 0.26) and RF (x¯ = 0.22). The overall RMSE in calibration models was lower in PCR models (X = 0.82) compared to L-SVM (x¯ = 0.86) and PLS (x¯ = 0.90). ENR, XGBoost and RF also had higher RMSE (x¯ = 0.90 - 0.92). Effective wavelengths selection using the interval partial least-squares regression (iPLS), improved the performance of the models but did not perform as good as the PLS. SNV pre-treatment was useful in improving model performance.

Prediction of fatty acid composition in intact and minced fat of European autochthonous pigs breeds by near infrared spectroscopy

The fatty acids profile has been playing a decisive role in recent years, thanks to technological, sensory and health demands from producers and consumers. The application of NIRS technique on fat tissues, could lead to more efficient, practical, and economical in the quality control. The study aim was to assess the accuracy of Fourier Transformed Near Infrared Spectroscopy technique to determine fatty acids composition in fat of 12 European local pig breeds. A total of 439 spectra of backfat were collected both in intact and minced tissue and then were analyzed using gas chromatographic analysis. Predictive equations were developed using the 80% of samples for the calibration, followed by full cross validation, and the remaining 20% for the external validation test. NIRS analysis of minced samples allowed a better response for fatty acid families, n6 PUFA, it is promising both for n3 PUFA quantification and for the screening (high, low value) of the major fatty acids. Intact fat prediction, although with a lower predictive ability, seems suitable for PUFA and n6 PUFA while for other families allows only a discrimination between high and low values.

A quantitative on-line analysis of salt in cured ham by near-infrared spectroscopy and chemometrics

In this work, non-invasive near-infrared spectroscopy (NIRS) combined with chemometrics was evaluated as a possible online analytical technique to categorize pieces of cured ham on the industrial production line based on their maximum sodium content. Stifle muscle was selected for the development of the NIRS prediction models because it is the one with the highest sodium content and the easiest in terms of accessibility for spectral measurement. In the study, samples with varying thicknesses were taken. The suitability of this method is demonstrated when a 5 mm sample is used for the construction of the model, obtaining the best fit with an R²cv of 92% and a prediction error of 0.11% sodium that coincides with the error of the reference method. In conclusion, a method is proposed for the direct determination of sodium content on the production line which allows the different pieces of ham to be quickly categorized according to their salt content.

Prediction of stable isotopes and fatty acids in subcutaneous fat of Iberian pigs by means of NIR: A comparison between benchtop and portable systems

The potential of a portable Near Infrared spectrophotometer compared with that of NIR benchtop equipment is assessed to determine the¹³C/¹²C relationship of stable isotopes and the fatty acid content. 105 samples of subcutaneous fat of Iberian pigs collected at the time of their slaughter have been analyzed. The analysis of stable isotopes and gas chromatography were the methods of reference used. The samples were analyzed without prior handling (portable and benchtop NIR) and after extracting the fat (benchtop NIR). The results show that with the portable equipment it is possible to determine δ¹³C (‰), 12 fatty acids, and 5 summations of fatty acids (SFA, MUFA, PUFA, w3, and w6), while with the benchtop NIR equipment it is possible to measure δ¹³C (‰), 16 fatty acids, and the 5 summationsof fatty acids. The correlation coefficients of the portable equipment were slightly lower than those of the NIR benchtop equipment.

Classification of polymorphic forms of fluconazole in pharmaceuticals by FT-IR and FT-NIR spectroscopy

The main goal of this work was to test the ability of vibrational spectroscopy techniques to differentiate between different polymorphic forms of fluconazole in pharmaceutical products. These are mostly manufactured with fluconazole as polymorphic form II and form III. These crystalline forms may undergo polymorphic transition during the manufacturing process or storage conditions. Therefore, it is important to have a method to monitor these changes to ensure the stability and efficacy of the drug. Each of FT-IR or FT-NIR spectra were associated to partial least squares-discriminant analysis (PLS-DA) for building classification models to distinguish between form II, form III and monohydrate form. The results has shown that combining either FT-IR or FT-NIR to PLS-DA has a high efficiency to classify various fluconazole polymorphs, with a high sensitivity and specificity. Finally, the selectivity of the PLS-DA models was tested by analyzing separately each of three following samples by FT-IR and FT-NIR: lactose monohydrate, which is an excipient mostly used for manufacturing fluconazole pharmaceutical products, itraconazole and miconazole. These two last compounds mimic potential contaminants and belong to the same class as fluconazole. Based on the plots of Hotelling's T² vs Q residuals, pure compounds of miconazole and itraconazole, that were analyzed separately, were significantly considered outliers and rejected. Furthermore, binary mixtures consist of fluconazole form-II and monohydrate form with different ratios were used to test the suitability of each technique FT-IR and FT-NIR with PLS-DA to detect minimum contaminant or polymorphic conversion from a polymorphic form to another using also the plots of Hotelling's T² vs Q residuals.

NIR Spectroscopy for Discriminating and Predicting the Sensory Profile of Dry-Cured Beef "Cecina"

For Protected Geographical Indication (PGI)-labeled products, such as the dry-cured beef meat "cecina de León", a sensory analysis is compulsory. However, this is a complex and time-consuming process. This study explores the viability of using near infrared spectroscopy (NIRS) together with artificial neural networks (ANN) for predicting sensory attributes. Spectra of 50 samples of cecina were recorded and 451 reflectance data were obtained. A feedforward multilayer perceptron ANN with 451 neurons in the input layer, a number of neurons varying between 1 and 30 in the hidden layer, and a single neuron in the output layer were optimized for each sensory parameter. The regression coefficient R squared (RSQ > 0.8 except for odor intensity) and mean squared error of prediction (MSEP) values obtained when comparing predicted and reference values showed that it is possible to predict accurately 23 out of 24 sensory parameters. Although only 3 sensory parameters showed significant differences between PGI and non-PGI samples, the optimized ANN architecture applied to NIR spectra achieved the correct classification of the 100% of the samples while the residual mean squares method (RMS-X) allowed 100% of non-PGI samples to be distinguished.

Prediction of Sensory Parameters of Cured Ham: A Study of the Viability of the Use of NIR Spectroscopy and Artificial Neural Networks

Dry-cured ham is a high-quality product owing to its organoleptic characteristics. Sensory analysis is an essential part of assessing its quality. However, sensory assessment is a laborious process which implies the availability of a trained tasting panel. The aim of this study was the prediction of dry-ham sensory characteristics by means of an instrumental technique. To do so, an artificial neural network (ANN) model for the prediction of sensory parameters of dry-cured hams based on NIR spectral information was developed and optimized. The NIR spectra were obtained with a fiber-optic probe applied directly to the ham sample. In order to achieve this objective, the neural network was designed using 28 sensory parameters analyzed by a trained panel for sensory profile analysis as output data. A total of 91 samples of dry-cured ham matured for 24 months were analyzed. The hams corresponded to two different breeds (Iberian and Iberian x Duroc) and two different feeding systems (feeding outdoors with acorns or feeding with concentrates). The training algorithm and ANN architecture (the number of neurons in the hidden layer) used for the training were optimized. The parameters of ANN architecture analyzed have been shown to have an effect on the prediction capacity of the network. The Levenberg–Marquardt training algorithm has been shown to be the most suitable for the application of an ANN to sensory parameters

Authentication of black cohosh (Actaea racemosa) dietary supplements based on chemometric evaluation of hydroxycinnamic acid esters and hydroxycinnamic acid amides

Ester and amide derivatives of hydroxycinnamic acids are found in black cohosh (Actaea racemosa) and other Actaea plants. These two compound groups were evaluated for authentication of black cohosh dietary supplements. The hydroxycinnamic acid esters (HCAE) were profiled by ultra-performance liquid chromatography-photodiode array detection (UPLC-PDA). The hydroxycinnamic acid amides (HCAA) were acquired simultaneously by mass spectrometry-multiple reaction monitoring (UPLC-MRM) mode. In contrast with the traditional HCAE method using 8 compounds, profiles of HCAA using only 4 feruloyl dopamine-O-hexosides was more convenient for peak by peak comparison. Partial least square discriminant analysis (PLS-DA) was applied to both HCAE and HCAA datasets. Authenticated plant samples of five Actaea species were randomly divided into training and test sets to build and validate the two PLS-DA models. Both models provided reasonable estimates for the classification of A. racemosa and other Actaea plant samples. However, HCAA model performs better in sensitivity, specificity, and accuracy. Assessment of supplement samples provided quite different results for the solid and liquid dietary supplement samples, indicating the dosage form could affect the composition of marker compounds. Graphical abstract.

Detection of cold chain breaks using partial least squares-class modelling based on biogenic amine profiles in tuna

The maintenance of the cold chain is essential to ensure foodstuff conformity and safety. However, gaps in the cold chain may be expected so designing analytical methods capable to detect cold chain breaks is a worthwhile issue. In this paper, the possibility of using the amount of nine biogenic amines (BAs) determined in Thunnus albacares by HPLC-FLD for detecting cold chain breaks is approached. Tuna is stored at 3 different temperature conditions for 8 storage periods. The evolution of the content of BAs is analyzed through parallel factor analysis (PARAFAC), in such a way that storage temperature, BAs and storage time profiles are estimated. PARAFAC has made it possible to observe two spoilage routes with different relative evolution of BAs. In addition, it has enabled to estimate the storage time, by considering the three storage temperatures, with errors of 0.5 and 1.0 days in fitting and in prediction, respectively. Furthermore, a class-modelling technique based on partial least squares is sequentially applied to decide, from the amount of BAs, if there has been a cold chain break. Firstly, samples stored at 25 °C are statistically discriminated from those kept at 4 °C and -18 °C; next, frozen samples are distinguished from those refrigerated. In the first case, the probabilities of false non-compliance and false compliance are almost zero, whereas in the second one, both probabilities are 10%. Globally, the results of this work have pointed out the feasibility of using the amount of BAs together with PLS-CM to decide if the cold chain has been maintained or not.

Assessing the influence of temperature on NIRS prediction models for the determination of sodium content in dry-cured ham slices

Temperature fluctuations are a key factor in the development of prediction models using near infrared spectroscopy (NIRS). In the present study, this influence has been investigated and a methodology has been proposed to reduce the effect of sample temperature on NIRS model prediction of the sodium content in dry-cured ham slices. Spectra were taken directly from the slices using a remote measurement probe (for non-contact analysis) at three different temperature ranges: -12 °C to -5°C, -5°C to 10 °C and 10 °C to 20 °C. Local and global temperature compensation methods were established. Partial-least squares (PLS) regression was used as a chemometrics tool to perform the calibrations. The results showed that local models were sensitive to changes in temperature, while a global temperature model using sample spectra over the entire temperature range showed good prediction ability, reducing the error caused by temperature fluctuations to acceptable levels for practical applications.

Application of the angle measure technique as image texture analysis method for the identification of uranium ore concentrate samples: New perspective in nuclear forensics

The identification of interdicted nuclear or radioactive materials requires the application of dedicated techniques. In this work, a new approach for characterizing powder of uranium ore concentrates (UOCs) is presented. It is based on image texture analysis and multivariate data modelling. 26 different UOCs samples were evaluated applying the Angle Measure Technique (AMT) algorithm to extract textural features on samples images acquired at 250× and 1000× magnification by Scanning Electron Microscope (SEM). At both magnifications, this method proved effective to classify the different types of UOC powder based on the surface characteristics that depend on particle size, homogeneity, and graininess and are related to the composition and processes used in the production facilities. Using the outcome data from the application of the AMT algorithm, the total explained variance was higher than 90% with Principal Component Analysis (PCA), while partial least square discriminant analysis (PLS-DA) applied only on the 14 black colour UOCs powder samples, allowed their classification only on the basis of their surface texture features (sensitivity>0.6; specificity>0.6). This preliminary study shows that this method was able to distinguish samples with similar composition, but obtained from different facilities. The mean angle spectral data obtained by the image texture analysis using the AMT algorithm can be considered as a specific fingerprint or signature of UOCs and could be used for nuclear forensic investigation.

Manufacturer identification and storage time determination of "Dong'e Ejiao" using near infrared spectroscopy and chemometrics

We have developed a set of chemometric methods to address two critical issues in quality control of a precious traditional Chinese medicine (TCM), Dong'e Ejiao (DEEJ). Based on near infrared (NIR) spectra of multiple samples, the genuine manufacturer of DEEJ, e.g. Dong'e Ejiao Co., Ltd., was accurately identified among 21 suppliers by the fingerprint method using Hotelling T(2), distance to Model X (DModX), and similarity match value (SMV) as discriminate criteria. Soft independent modeling of the class analogy algorithm led to a misjudgment ratio of 6.2%, suggesting that the fingerprint method is more suitable for manufacturer identification. For another important feature related to clinical efficacy of DEEJ, storage time, the partial least squares-discriminant analysis (PLS-DA) method was applied with a satisfactory misjudgment ratio (15.6%) and individual prediction error around 1 year. Our results demonstrate that NIR spectra comprehensively reflect the essential quality information of DEEJ, and with the aid of proper chemometric algorithms, it is able to identify genuine manufacturer and determine accurate storage time. The overall results indicate the promising potential of NIR spectroscopy as an effective quality control tool for DEEJ and other precious TCM products.

Detection of adulteration in hydrated ethyl alcohol fuel using infrared spectroscopy and supervised pattern recognition methods

This paper proposes an analytical method to detect adulteration of hydrated ethyl alcohol fuel based on near infrared (NIR) and middle infrared (MIR) spectroscopies associated with supervised pattern recognition methods. For this purpose, linear discriminant analysis (LDA) was employed to build a classification model on the basis of a reduced subset of wavenumbers. For variable selection, three techniques are considered, namely the successive projection algorithm (SPA), the genetic algorithm (GA) and a stepwise formulation (SW). For comparison, models based on partial least squares discriminant analysis (PLS-DA) were also employed using full-spectrum. The method was validated in a case study involving the classification of 181 hydrated ethyl alcohol fuel samples, which were divided into three different classes: (1) authentic samples; (2) samples adulterated with water and (3) samples contaminated with methanol. LDA/GA and PLS-DA models were found to be the best methods for classifying the spectral data obtained in NIR region, which achieved a correct prediction rate of 100% in the test set, while the LDA/SPA and LDA/SW were correctly classified at 84.4% and 97.8%, respectively. For MIR data, all models (PLS-DA and LDA coupled with the SW, SPA and GA) employed in this study correctly classified all samples in the test set.

Two class-modelling techniques that give families of class-models and their relation with the structure of the data

In class-modelling problems, which are again becoming increasingly important, there are two parameters to value the quality of the class-model built for a category, namely sensitivity and specificity. Using them as criteria, in this paper, two different approaches to class-modelling problems are presented, approaches that differ from other usual methods in the fact that they provide not just one class-model per category but a set of different class-models that accounts for the possible pairs of sensitivity-specificity values attainable for a given data set. One of the proposals is partial least squares class-modelling (PLS-CM) that, by the joint use of PLS with binary responses and the posterior statistical modelling of the distribution of the computed responses, permits the estimation of the risks related to the decision of assigning a sample into a class, and thus, the values of sensitivity and specificity. The other proposed method, Pareto-optimal front in class-modelling, is an analytical approach posed in a multi-response optimization framework, the one that corresponds to trying to simultaneously maximise the sensitivity and specificity of a class-model. Additionally, the whole family of computed class-models is validated in prediction by using cross-validation, showing the stability of both methods for prediction. The case-studies show the complementariness of both approaches and, in particular, that the joint use of both techniques allows the user to detect possible structures in the data set especially inadequate for PLS. The results, i.e. the whole set of sensitivity-specificity values achievable for a given problem, are graphically represented to improve its study and make it easy to make a decision about the model.

Machine learning based prediction for peptide drift times in ion mobility spectrometry

MOTIVATION

Ion mobility spectrometry (IMS) has gained significant traction over the past few years for rapid, high-resolution separations of analytes based upon gas-phase ion structure, with significant potential impacts in the field of proteomic analysis. IMS coupled with mass spectrometry (MS) affords multiple improvements over traditional proteomics techniques, such as in the elucidation of secondary structure information, identification of post-translational modifications, as well as higher identification rates with reduced experiment times. The high throughput nature of this technique benefits from accurate calculation of cross sections, mobilities and associated drift times of peptides, thereby enhancing downstream data analysis. Here, we present a model that uses physicochemical properties of peptides to accurately predict a peptide's drift time directly from its amino acid sequence. This model is used in conjunction with two mathematical techniques, a partial least squares regression and a support vector regression setting.

RESULTS

When tested on an experimentally created high confidence database of 8675 peptide sequences with measured drift times, both techniques statistically significantly outperform the intrinsic size parameters-based calculations, the currently held practice in the field, on all charge states (+2, +3 and +4).

AVAILABILITY

The software executable, imPredict, is available for download from http:/omics.pnl.gov/software/imPredict.php

CONTACT

rds@pnl.gov

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Meat quality assessment using biophysical methods related to meat structure

This paper overviews the biophysical methods developed to gain access to meat structure information. The meat industry needs reliable meat quality information throughout the production process in order to guarantee high-quality meat products for consumers. Fast and non-invasive sensors will shortly be deployed, based on the development of biophysical methods for assessing meat structure. Reliable meat quality information (tenderness, flavour, juiciness, colour) can be provided by a number of different meat structure assessment either by means of mechanical (i.e., Warner-Bratzler shear force), optical (colour measurements, fluorescence) electrical probing or using ultrasonic measurements, electromagnetic waves, NMR, NIR, and so on. These measurements are often used to construct meat structure images that are fusioned and then processed via multi-image analysis, which needs appropriate processing methods. Quality traits related to mechanical properties are often better assessed by methods that take into account the natural anisotropy of meat due to its relatively linear myofibrillar structure. Biophysical methods of assessment can either measure meat component properties directly, or calculate them indirectly by using obvious correlations between one or several biophysical measurements and meat component properties. Taking these calculations and modelling the main relevant biophysical properties involved can help to improve our understanding of meat properties and thus of eating quality.

Noncontact analysis of the fiber weight per unit area in prepreg by near-infrared spectroscopy

The fiber weight per unit area in prepreg is an important factor to ensure the quality of the composite products. Near-infrared spectroscopy (NIRS) technology together with a noncontact reflectance sources has been applied for quality analysis of the fiber weight per unit area. The range of the unit area fiber weight was 13.39-14.14mgcm(-2). The regression method was employed by partial least squares (PLS) and principal components regression (PCR). The calibration model was developed by 55 samples to determine the fiber weight per unit area in prepreg. The determination coefficient (R(2)), root mean square error of calibration (RMSEC) and root mean square error of prediction (RMSEP) were 0.82, 0.092, 0.099, respectively. The predicted values of the fiber weight per unit area in prepreg measured by NIRS technology were comparable to the values obtained by the reference method. For this technology, the noncontact reflectance sources focused directly on the sample with neither previous treatment nor manipulation. The results of the paired t-test revealed that there was no significant difference between the NIR method and the reference method. Besides, the prepreg could be analyzed one time within 20s without sample destruction.

Supervised pattern recognition in food analysis

Data analysis has become a fundamental task in analytical chemistry due to the great quantity of analytical information provided by modern analytical instruments. Supervised pattern recognition aims to establish a classification model based on experimental data in order to assign unknown samples to a previously defined sample class based on its pattern of measured features. The basis of the supervised pattern recognition techniques mostly used in food analysis are reviewed, making special emphasis on the practical requirements of the measured data and discussing common misconceptions and errors that might arise. Applications of supervised pattern recognition in the field of food chemistry appearing in bibliography in the last two years are also reviewed.

Rapid determination of sulfonamides in milk samples using fluorescence spectroscopy and class modeling with n-way partial least squares

In this paper, a methodology to evaluate the probability of false non-compliance and false compliance for screening methods, which give first or second-order multivariate signals is proposed. For this task 120 samples of 6 different kinds of milk have been measured by excitation-emission fluorescence. The samples have been spiked with different amounts of three sulfonamides (sulfadiazine, sulfamerazine and sulfamethazine). These substances have been classified in group B1 (veterinary medicines and contaminants) of annex I of Directive 96/23/EC. The European Union (Commission Regulation EC no. 281/96) has set the maximum residue level (MRL) of total sulfonamides at 100 microg kg(-1) in muscle, liver, kidney and milk. The work shows that excitation-emission fluorescence together with the partial least squares class modeling (PLS-CM) procedure may be a suitable and cheap screening method for the total amount of sulfonamides in milk. Three models, PLS-CM, have been built, for the emission and excitation spectra (first-order signals) and for the excitation-emission matrices (second-order signals). In all the cases it reaches probabilities of false compliance below 5% as required by Decision 2002/657/EC. With the same flourescence signals, the total quantity of sulfonamide was calibrated using 2-PLS, 3-PLS and PARAFAC regressions. Using this quantitative approach, the capability of detection, CCbeta, around the MRL has been estimated between 114.3 and 115.1 microg kg(-1) for a probability of false non-compliance and false compliance equal to 5%.