Modified Partial Least Squares Regression Research Articles

Prediction of the Fatty Acid Profiles of Iberian Pig Products by Near Infrared Spectroscopy: A Comparison Between Multiple Regression Tools and Artificial Neural Networks

AbstractIn this study, the feasibility of predicting the lipid profiles of Iberian ham and shoulder samples by using near infrared (NIR) spectroscopy was evaluated. Gas chromatography analysis was the reference method used. The muscles analyzed and recorded by NIR spectroscopy were 76 Biceps femoris for Iberian hams and 72 Brachiocephalicus for Iberian shoulders. NIR calibrations were carried out by using two methods: modified partial least squares regression (MPLS) and artificial neural networks (ANN). With the MPLS method, it was possible to obtain equations with regression’s coefficients (RSQ) of > 0.5 for 5 individual fatty acids and 3 summations: polyunsaturated fatty acids, n3 and n6. The use of neural networks made it possible to find equations with RSQ of > 0.5 for 10 individual fatty acids, all of which are present in over 90% of the samples, and 5 summations of saturated, monounsaturated, and polyunsaturated fatty acids (SFA, MUFA, PUFA), n3 and n6, finding that the calibration curves of the fatty acids C18:1, C18:2n6, and C18:3n3 presented RSQ’s of > 0.7. The results obtained indicate that NIR spectroscopy could be a very useful technology for the quality control of cured products as it allows estimating the main fatty constituents quickly and without using reagents.

Assessing the Application of Near-Infrared Spectroscopy to Determine Saccharification Efficiency of Corn Biomass

AbstractNowadays, in the bioethanol production process, improving the simplicity and yield of cell wall saccharification procedure represent the main technical hurdles to overcome. This work evaluated the application of a rapid and cost-effective technology such as near -infrared spectroscopy (NIRS) for easily predict saccharification efficiency from corn stover biomass. Calibration process focussing on the number of samples and the genetic background of the maize inbred lines were tested; while Modified Partial Least Squares Regression (MPLS) and Multiple Linear Regression (MLR) were assessed in predictions. The predictive capacity of the NIRS models was mainly determined by the coefficient of determination (r2ev) and the index of prediction to deviation (RPDev) in external validation. Overall, we could check a better efficiency of the NIRS calibration process for saccharification using larger number of observations (1500 sample set) and genetic backgrounds; while MPLS regression provided better prediction statistics (r2ev = 0.80; RPDev = 2.21) compared to MLR (r2ev = 0.68; RPDev = 1.75). These results indicate that NIRS could be successfully implemented as a large-phenotyping tool in order to test the saccharification potential of corn biomass.

Integrative quantitative and qualitative analysis for the quality evaluation and monitoring of Danshen medicines from different sources using HPLC-DAD and NIR combined with chemometrics.

The root and rhizome of Salvia miltiorrhiza (Danshen in short) is a well-known herbal medicine used to treat cardiovascular diseases in the world. In China, the roots and rhizomes of several other Salvia species (Non-Danshen in short) are also used as this medicine in traditional folk medicine by local herbalists. Differences have been reported in these medicines originating from different sources, and their quality variation needs to be clearly investigated for effective clinical application. This study presented a comprehensive quality evaluation and monitoring for Danshen from 27 sampling sites and Non-Danshen from other 5 Salvia species based on a high-performance liquid chromatography-diode array detector (HPLC-DAD) and near-infrared (NIR), with the combination of chemometric models. The results showed that cryptotanshinone, tanshinone IIA, tanshinone I, salvianolic acid B, salvianic acid A sodium, dihydrotanshinone I, and rosmarinic acid in these medicines from different sources exhibited great variations. Referring to the standards in Chinese Pharmacopoeia (CP), European Pharmacopeia (EP), and United States Pharmacopeia (USP), Non-Danshen from S. brachyloma, S. castanea, S. trijuga, S. bowleyana, and S. przewalskii were assessed as unqualified, and Danshen in the Shandong Province had the best quality due to the high qualified rate. Based on random forest (RF) and partial least-squares discriminant analysis (PLS-DA), NIR technique could successfully monitor the quality of these medicines by discriminating the species and regions with the accuracies of 100.00 and 99.60%, respectively. Additionally, modified partial least-squares regression (MPLSR) models were successfully constructed to investigate the feasibility of NIR fingerprints for the prediction of the quality indicators in these medicines. The optimized models obtained the best results for the total of tanshinone IIA, tanshinone I, and cryptotanshinone (TTC), tanshinone IIA, and salvianolic acid B, with the relative prediction deviation (RPD) of 4.08, 3.92, and 2.46, respectively. In summary, this study demonstrated that HPLC-DAD and NIR techniques can complement each other and could be simultaneously applied for evaluating and monitoring the quality of Danshen medicines.

Non-destructive determination of grass pea and pea flour adulteration in chickpea flour using near-infrared reflectance spectroscopy and chemometrics.

In order to obtain more economic gains, some food products are adulterated with low-cost substances, if they are toxic, they may pose public health risks. This has called forth the development of quick and non-destructive methods for detection of adulterants in food. Near-infrared reflectance spectroscopy (NIRS) has become a promising tool to detect adulteration in various commodities. We have developed rapid NIRS based analytical methods for quantification of two cheap adulterants (grass pea and pea flour) in a popular Indian food material, chickpea flour. The NIRS spectra of pure chickpea, pure grass pea, pure pea flour and adulterated samples of chickpea flour with grass pea and pea flour (1-90%) (w/w) were acquired and preprocessed. Calibration models were built based on modified partial least squares regression (MPLSR), partial least squares (PLS), principal component regression (PCR) methods. Based on lowest values of standard error of calibration (SEC) and standard error of cross-validation (SECV), MPLSR-NIRS models were selected. These models exhibited coefficient of determination (R2 ) of 0.999, 0.999, SEC of 0.905, 0.827 and SECV of 1.473, 1.491 for grass pea and pea, respectively. External validation revealed R2 and standard error of prediction (SEP) of 0.999 and 1.184, 0.997 and 1.893 for grass pea and pea flour, respectively. The statistics confirmed that our MPLSR-NIRS based methods are quite robust and applicable to detect grass pea and pea flour adulterants in chickpea flour samples and have potential for use in detecting food fraud. © 2022 Society of Chemical Industry.

Prediction of maize flour adulteration in chickpea flour (besan) using near infrared spectroscopy

The present study was performed to develop Near-infrared spectroscopy based prediction method for the quantification of the maize flour adulteration in chickpea flour. Adulterated samples of Chickpea flour (besan) were prepared by spiking different concentrations of maize flour with pure Chickpea flour in the range of 1–90% (w/w). The spectra of pure Chickpea flour, pure maize flour, and adulterated samples of Chickpea flour with maize flour were acquired as the logarithm of reciprocal of reflectance (log 1/R) in the entire Visible-NIR wavelength range of 400–2498 nm. The acquired spectra were pre-processed by Ist derivative, standard normal variate, and detrending. The calibration models were developed using modified partial least square regression (MPLSR), partial least square regression and principal component regression. The optimal model was selected on the basis of highest values of the coefficient of determination (RSQ), one minus variance ratio (1-VR) and lowest values of standard errors of calibration (SEC), and standard error of cross-validation (SECV). MPLSR model having RSQ and 1-VR value of 0.999 and 0.996 having SEC and SECV value of 1.092 and 2.042 was developed for quantification of maize flour adulteration in chickpea flour. Cross validation and external validation of the developed models resulted in RSQ of 0.999, 0.997 and standard error of prediction of 1.117, and 2.075, respectively.Supplementary InformationThe online version contains supplementary material available at 10.1007/s13197-022-05456-7.

Near-Infrared Spectroscopy to Predict Provitamin A Carotenoids Content in Maize

Vitamin A deficiency (VAD) is a public health issue worldwide. Provitamin A (PVA) biofortified maize serves as an alternative to help combat VAD. Breeding efforts to develop maize varieties with high PVA carotenoid content combine molecular and phenotypic selection strategies. The phenotypic assessment of carotenoids is currently done using liquid chromatography, a precise but time- and resource-consuming methodology. Using near-infrared spectroscopy (NIRS) could increase the breeding efficiency. This study used ultra-performance liquid chromatography (UPLC) data from 1857 tropical maize genotypes as a training set and NIRS data to do an independent test of a set of 650 genotypes to predict PVA carotenoids using Bayesian and modified partial least square (MPLS) regression models. Both regression methods produced similar prediction accuracies for the total carotenoids (r2 = 0.75), lutein (r2 = 0.55), zeaxanthin (r2= 0.61), β-carotene (r2 = 0.22) and β-cryptoxanthin (BCX) (r2 = 0.57). These results demonstrate that Bayesian and MPLS regression of BCX on NIRS data can be used to predict BCX content, the current focus on PVA enhancement, and thus offers opportunities for high-throughput phenotyping at a low cost, especially in the early stages of PVA maize breeding pipeline when many genotypes must be screened.

Feasibility study on the use of ATR-FTIR spectroscopy as a tool for the estimation of wine polysaccharides

Polysaccharides are the main macromolecules of colloidal nature in wines. These compounds play a critical role in stabilizing other molecules in solution and thus modifying the wine processing and organoleptic properties. Different analytical techniques have been proposed for their determination. However, most of them are complicated and time-consuming. To overcome these drawbacks, Fourier transform infrared spectroscopy (FTIR) has been evaluated in this study for the estimation of wine polysaccharides in a fast and non-destructive way.Spectral data have been correlated with wine polysaccharide contents by modified partial least squares regression (MPLS) using different spectral pretreatments. MPLS models developed have revealed the potential of FTIR analysis for the routine screening of polysaccharides rich in arabinose and galactose (PRAG), rhamnogalacturonans types II (RG-II), mannoproteins (MP) and total soluble polysaccharides (TSP) in wine samples, obtaining standard errors of prediction from 6.07 to 8.44%. Monitoring the wine polysaccharides can assist in the elaboration of the wines according to their requirements and improving quality to satisfy consumer preferences.

An Modified PLSR Method in Prediction

Among many statistical methods for linear models with the multicollinearity problem, partial least squares regression (PLSR) has become, in recent years, increasingly popular and, very often, the best choice. However, while dealing with the predicting problem from automobile market, we noticed that the results from PLSR appear unstable though it is still the best among some standard statistical methods. This unstable feature is likely due to the impact of the information contained in explanatory variables that is irrelevant to the response variable. Based on the algorithm of PLSR, this paper introduces a new method, modified partial least squares regression (MPLSR), to emphasize the impact of the relevant information of explanatory variables on the response variable. With the MPLSR method, satisfactory predicting results are obtained in the above practical problem. The performance of MPLSR, PLSR and some standard statistical methods are compared by a set of Monte Carlo experiments. This paper shows that the MPLSR is the most stable and accurate method, especially when the ratio of the number of observation and the number of explanatory variables is low.

Fermentation quality evaluation of tea by estimating total catechins and theanine using near-infrared spectroscopy

Concentrations of total catechins and theanine are important indices for monitoring the degree of fermentation, a key factor in determining the types, qualities, and taste characteristics of tea, during manufacturing process. Until now, there has not been any near-infrared (NIR) study applying theanine as an index for monitoring the degree of fermentation of tea. It is necessary to understand and monitor the degree of fermentation according to the concentrations of total catechins and theanine in tea by using NIR spectroscopy, especially for manufacturers of tea products. NIR spectroscopy integrated with modified partial least-squares regression (MPLSR) method was used for spectral analysis of total catechins and theanine in 161 tea samples in the full visible NIR wavelength range of 400−2498 nm (25000−4003 cm−1). Both the optimal calibration models of total catechins and theanine built by MPLSR showed satisfactory predictability, with coefficient of determination of calibration (Rc2) and residual predictive deviation (RPD) up to 0.94 and 3.38, respectively. High total accuracy rates of 100 % were also achieved for the discrimination of green tea, partially fermented tea, and black tea in the validation sets by the optimal MPLSR calibration model of theanine. The results indicated that NIR spectroscopy has the potential to be adopted as an effective method of rapid and accurate inspection of the degree of fermentation of tea. This technique could contribute substantially to quality assurance during fermentation process of tea.

Comparison of artificial neural networks and multiple regression tools applied to near infrared spectroscopy for predicting sensory properties of products from quality labels

In products from quality labels a sensory analysis is obligatory although this is a slow and expensive process. This study examines the prediction of the sensory parameters of chorizo dry-cured sausage by using NIRS technology and the application of chemometric methods such as MPLS (Modified Partial Least Square regression) and ANN (Artificial Neural Networks). The results show that by applying ANN it is possible to predict the 20 sensory parameters analyzed with RSQ values of from 0.61 to 0.92; these values are always higher than those obtained by prediction using MPLS. Moreover, the combination of NIRS and RMS-X residual discrimination allowed the correct classification of 94.4% of the samples according to whether or not they belonged to a certain Quality Label.

Procjena količine makro- i mikroelemenata u hrvatskim tradicijskim kultivarima graha (Phaseolus vulgaris L.) korištenjem spektroskopije u bliskome infracrvenom području (NIRS)

In this study, near-infrared spectroscopy (NIRS) was used to predict the contents of essential macro- and microelements in common bean (Phaseolus vulgaris L.) accessions of most widespread Croatian landraces. Total of 175 samples were used for the model development by modified partial least square (MPLS), principal component regression (PCR) and partial least square (PLS) techniques. Based on the coefficients of determination (R2), standard error of calibration (SEC) and error of prediction (SEP) the models developed were (i) nearly applicable for nitrogen (N) (0.89, 0.12 and 0.45 respectively), (ii) poor for iron (Fe), cinc (Zn), potassium oxide (K2O) and potassium (K), (iii) usable for phosphorus pentoxide (P2O5), phosphorus (P), phytic acid (PA) and manganese (Mn). The MPLS regression statistics suggested the most accurate models developed comparing with PLS and PCR. It was concluded that a wider set of common bean samples needs to be used for macro- and microelements prediction by NIRS.

Screening of Wine Extractable Total Phenolic and Ellagitannin Contents in Revalorized Cooperage By-products: Evaluation by Micro-NIRS Technology

Determining phenolic compounds of wood and its extractability to the hydroalcoholic medium is important in the oenological industry. The method proposed in this study copes with this issue in an in situ, non-destructive, and fast way. For this purpose, a number of oak by-product samples spectrally representative have been selected. Selected spectral data have been correlated with oak wood extractable polyphenols (extractable total phenolic content and extractable ellagitannin content) by modified partial least squares regression (MPLS) obtaining coefficients of determination (RSQ) greater than 0.9 and standard errors of prediction (SEP) between 13.68 and 23.51% for all parameters evaluated. The obtained results are comparable with those obtained using bench-top devices and present the advantage of its eventual friendly use out of lab. Development of applicable models in situ will allow a greater versatility and efficiency for the decision-making in the winemaking process on the adequacy and/or dosage of these by-products according to the requirements of the wine. The use of cooperage by-products as the source of copigments for wine leads to a sustainable and competitive cooperage industry, through waste reduction and by-product valorization.

Evaluation of Carbohydrate Concentrations in Phalaenopsis Using Near-infrared Spectroscopy

Carbohydrate concentrations are important indicators of the internal quality of Phalaenopsis. In this study, near-infrared (NIR) spectroscopy was used for quantitative analyses of fructose, glucose, sucrose, and starch in Phalaenopsis plants. Both modified partial least-squares regression (MPLSR) and stepwise multiple linear regression (SMLR) methods were used for spectral analysis of 302 Phalaenopsis samples in the full visible NIR wavelength range (400–2498 nm). Calibration models built by MPLSR were better than those built by SMLR. For fructose, the smoothed first derivative MPLSR model provided the best results, with a correlation coefficient of calibration (Rc) of 0.96, standard error of calibration (SEC) of 0.22% dry weight (DW), standard error of validation (SEV) of 0.28% DW, and bias of -0.01% DW. For glucose, the MPLSR model based on the smoothed first derivative spectra was the best (Rc = 0.96; SEC = 0.26% DW; SEV = 0.32% DW; and bias = 0.01% DW). The best MPLSR model of sucrose was developed using the smoothed first derivative spectra (Rc = 0.96; SEC = 0.24% DW; SEV = 0.31% DW; bias = -0.03% DW). Regarding starch, the smoothed first derivative MPLSR model showed the best effects (Rc = 0.91; SEC = 0.47% DW; SEV = 0.56% DW; bias = -0.02% DW). Both the MPLSR and SMLR models showed satisfactory predictability, indicating that NIR has the potential to be adopted as an effective method of rapid and accurate inspection of the carbohydrate concentrations of Phalaenopsis plants. This technique could contribute substantially to quality management of Phalaenopsis.

Feasibility study on the use of a portable micro near infrared spectroscopy device for the “in vineyard” screening of extractable polyphenols in red grape skins

There is substantial variation in levels of extractable phenolic compounds of red grapes (Vitis vinifera L.). Therefore, it could be desirable to known the aforesaid parameter at least for each vine. Nowadays, interest has shifted toward the development of portable vis/NIR systems, innovation in optical system design and miniaturization for its friendly use directly in the field.Spectra of intact grapes and grapes skins were recorded at harvest time in two different vintages (2016 and 2017 respectively) using a portable micro NIR spectrophotometer (908–1676 nm). A number of chemometric approaches have been used for spectral interrogation and evaluation of the aforesaid device. Spectral data have been correlated with red grape skin extractable polyphenols (total phenolic, anthocyanins and flavanols) by modified partial least squares regression (MPLS) using a number of spectral pretreatments. Moreover, different statistics strategies have been performed to develop a qualitative analysis of the data (linear discriminant analysis, discriminant partial least square analyses and Pearson's similarity index).After an exhaustive analysis of the obtained results in two different seasons, it can be concluded that the use of the portable micro NIR device for the “in vineyard” screening of extractable polyphenols in red grape skins is hampered by a number of factors. Environmental and physiological conditions should be considered to evaluate and remove factors that hamper a good sorting the berries according to their extractable polyphenol contents.

Evaluation of Water Potentials of Leafy Vegetables Using Hyperspectral Imaging

To effectively and precisely manage crop production is important to meet increasing demands of high quality and safe agricultural products. In this study, evaluation of water stress in terms of water potentials of leafy vegetables using hyperspectral imaging technology was conducted for the future development of an agri-robotic system in greenhouses. The leafy vegetable Fengjing Pakchoi was cultivated under water stress treatments in a phytotron; both water potential and hyperspectral images were measured. The Modified Partial Least Squares Regression (MPLSR) with a correlation coefficient of 0.826 was successfully developed to determine the water potentials of leaves using hyperspectral imaging. Uneven distribution of water potentials on leaves was illustrated and discussed. Water stress which reflected irrigation status was visualized and compared by pseudocolor contour figures. This study had established a basis for the subsequent development of an agri-robotic system for automation and precision irrigation in greenhouses.

Determination of fibre and protein content in heterogeneous pastures using field spectroscopy and ultrasonic sward height measurements

Feeding of livestock on pastures requires constant monitoring of diet composition to ensure consistent levels of animal production. The widely used but conventional techniques to measure the components of feed are impractical to obtain in-field forage quality status for making real-time decisions. Assessment of forage quality parameters using proximal sensing is of particular interest. The present study aimed to demonstrate the potential of using a combination of ultrasonic and canopy reflectance data to predict forage quality variables of heterogeneous pastures. A field experiment with pastures continuously grazed by cows with three stocking density treatments (moderate, lenient and very lenient stocking) was used to calibrate ultrasonic and hyperspectral reflectance sensors. Hyperspectral analysis by a modified partial least square regression (MPLSR) resulted in maximum accuracy for the prediction of acid detergent fibre (ADF) and crude protein (CP) (R2calibration=0.63–0.85). Any reduction of hyperspectral data into vegetation indices based on few specific narrow wavebands or satellite broadbands reduced prediction accuracy significantly. However, prediction of ADF and CP was improved by a combined analysis of ultrasonic sward height and vegetation indices or satellite broadbands, so that most calibration models exceeded an RPD (ratio of standard deviation and standard error of prediction) value of 1.4, which is considered as an acceptable predicting capability for variable field condition. Our findings showed that combined sensing systems using reflectance and ultrasonic sensors may provide acceptable prediction accuracies for practical application even under extremely heterogeneous pasture conditions.

Partial least squares with a regularized weight

The key issue in partial least squares (PLS) is the computation of loading weight which describes the relations between the responses and observation variables. In this paper, we propose a simple modified PLS regression method to adaptively compute the loading weight. The weight is solved according to a regularization framework with a similarity constraint to keep the spectra–concentrate relations and meanwhile to penalty the variables with a large distance to the response. The regularized weights can accurately reflect the spectra–concentrate relations, and lead to a good-fitting and parsimonious model which is less sensitive to the latent variables. The proposed method is compared, in terms of root mean square errors of prediction, correlation coefficient and cumulative explained variance, to PLS on simulated and real near-infrared spectra data sets. Experimental results demonstrate the efficacy and effectiveness of the proposed method.

Optimization of Near‐Infrared Reflectance Model in Measuring Gelatinization Characteristics of Rice Flour with a Rapid Viscosity Analyzer (RVA) and Differential Scanning Calorimeter (DSC)

This study compared the calibration models generated by combinations of different mathematical and preprocessing treatments as well as regression algorithms to optimize the analysis of gelatinization properties of rice flour by using near‐infrared spectroscopy, in comparison with conventional techniques of differential scanning calorimetry (DSC) and rapid viscosity analysis (RVA). A total of 220 milled rice flours were used for model construction. A model generated by the modified partial least squares regression (MPLS) with mathematical treatment “2, 8, 8, 2” (second‐order derivative computed based on eight data points, and eight and two data points in the second smoothing, respectively) and detrend preprocessing was identified as the best for simultaneously measuring onset temperature (To), peak temperature (Tp), and conclusion temperature (Tc) of DSC. MPLS/“2, 8, 8, 2”/weighted multiplicative scattering correction preprocessing was identified as the best for RVA properties. The results indicated that near‐infrared reflectance spectroscopy could be used to rapidly predict gelatinization properties of rice flour for the purposes of quality evaluation of germplasm and selection of intermediate lines in breeding programs.

Prediction models based on multivariate statistical methods and their applications for predicting railway freight volume

Four prediction models based on the multivariate statistical methods are constructed in this work and they are successfully applied in predicting the Railway Freight Volume (RFV). RFV directly reflects the regional economic states such as production improvement and economic restructuring. Accurately predicting the RFV is of great use in production planning, decision making, labor allocating, etc. In this work, based on the multivariate statistical methods, i.e. ordinary least squares regression (OLSR), principal component regression (PCR), partial least squares regression (PLSR), and modified partial least squares regression (MPLSR), four RFV prediction models are constructed and the detailed comparison is made by implementing them on a practical dataset. From the simulation results, the conclusion can be derived that the MPLSR based prediction model outperforms the other three models.

Studies on 5 Protein Fractions Prediction of Forage Legume Mixture by NIRS

This study was conducted to assess the feasibility of near-infrared reflectance spectroscopy (NIRS) as a rapid and reliable method for the estimation of crude protein (CP) fractions in forage legume mixtures (sudangrass and pea mixture, and kidney bean and potato mixture). A total of 178 samples were collected and their spectral reflectance obtained in the range of 400~2,500 nm. Of these, 50 samples were selected for calibration and validation, and 35 samples were used for calibration of the data set, and the modified partial least square regression (MPLSR) analysis was performed. The correlation coefficient (<TEX>$r^2$</TEX>) and the standard error of cross-validation (SECV) of the calibration models in the CP fractions, A, B1, B2, B3, and C, were 0.94 (1.05), 0.92 (0.74), 0.96 (0.95), 0.91 (0.42), and 0.83 (0.38), respectively. Fifteen samples were used for equation validation, and the <TEX>$r^2$</TEX> and the standard error of prediction (SEP) were 0.87 (1.45), 0.91 (0.49), 0.94 (1.13), 0.36 (0.96), and 0.74 (0.67), respectively. This study showed that NIRS could be an effective tool for the rapid and precise estimation of CP fractions in forage legume mixtures.