Determination of the superficial citral content on microparticles: An application of NIR spectroscopy coupled with chemometric tools

Abstract

This work evaluates near-infrared ( NIR ) spectroscopy coupled with chemometric tools for determining the superficial content of citral ( ) on microparticles. To perform this evaluation, using spray drying, citral was encapsulated in a matrix of dextrin using twelve combinations of citral:dextrin ratios ( CDR ) and inlet air temperatures ( IAT ). From each treatment, six samples were extracted, and their and NIR absorption spectral profiles were measured. Then, the spectral profiles, pretreated and randomly divided into modeling and validation datasets, were used to build the following prediction models: principal component analysis-multilinear regression ( PCA-MLR ), principal component analysis-artificial neural network ( PCA-ANN ), partial least squares regression ( PLSR ) and an artificial neural network ( ANN ). During the validation stage, the models showed values from 0.73 to 0.96 and a root mean squared error ( RMSE ) range of [0.061–0.140]. Moreover, when the models were compared, the full and optimized ANN models showed the best fits. According to this study, NIR coupled with chemometric tools has the potential for application in determining on microparticles, particularly when using ANN models.

Determination of mechanical properties of whey protein films during accelerated aging: Application of FTIR profiles and chemometric tools

“…Figure 3 shows that four PCs can explain variance in 96.03% for TS, and 98.06% for EB and EM respectively, being two enough to explain variance near to 90% for all properties. These findings are comparable to those reported by Yoplac et al (2019), four PCs explained over 99.5% of variance to determine citral in a matrix of dextrin; Nallan Chakravartula et al (2019) that using one PC explained variance in 79 and 94% to differentiate bread samples during drying; Sharma, Nani, and Kumar (2019) with three PCs explained variance in 98% to identified plasticizers in cling films, or Du et al (2020) who reached values above 90% to explain variance for shearing and bending strength of tea stem. In all the cases with four or less PCs were enough to explain variance in evaluated properties over 90%.…”

“…In all the cases the R 2 varied from 0.0022 to 0.205 (Table 2); these results show that this kind of models produce a very low fit for all the objective variables. A similar trend can be founding the results reported by Yoplac et al (2019) who use PCA‐MLR for NIR spectra to adjust around 0.75 to predict citral content in films, significantly lower than the other models tested in that research. Likewise, according to Castro et al (2020), these models can not differentiate between variables according to their predictive capacity, oversizing variables and as a consequence introducing noise to the prediction.…”

“…These results showed a high dispersion for TS, which, according to Esposito, Chin, Henseler, and Wang (2010), may be due to the overlapping and high collinearity of spectral profiles. In this sense the use of β‐coefficient, as a measure to weight the relative importance of each variable, improves models fit excluding noisy variables (Castro et al, 2020); this was used in the works of Yoplac et al (2019), Vásquez et al (2018), and Castro et al (2018) among others with relative success.…”

“…Similar results were presented by Vásquez et al (2018), who created an optimized PLSR model to know the hardness of cheese, reporting R 2 values of 0.8735; Du et al (2020) predicting bending and shearing strength for tea steam obtain R 2 values of 0.9939, 0.8045, and RMSE values of 0.9472, and 1.0770 respectively; Nallan Chakravartula et al (2019) predicting the drying time of edible coating on bread at two temperatures reported R 2 over 0.90 and RMSE around 2.5; Yoplac et al (2019) determining citral in a matrix of dextrin showed R 2 and RMSE values of 0.932 and 0.0770; or Castro et al (2020) that uses PLSR models to predict water activity on Quinoa grains reporting R 2 and RMSE values of 0.7760 and 0.0090 respectively. In all previously commented cases, the feasibility to use PLSR models was defined using R 2 and RMSE in similar way to our research.…”

“…These, in fact, enable us to extract maximum information from spectral profiles and correlate it with the material properties in order modeling its relationships. Among them, some of the most commonly used technique for prediction studies are Principal Component Analysis (PCA) (S. Liu, Kokot, & Will, 2009; Nallan Chakravartula et al, 2019; Rohman et al, 2011; Yoplac, Avila‐George, Vargas, Robert, & Castro, 2019), Multilinear Regression (MLR) and Partial Least Squares Regression (PLSR) (Du et al, 2020; S. Liu et al, 2009; Mauricio‐Iglesias et al, 2009; Nallan Chakravartula et al, 2019; Nunes et al, 2016; Rohman et al, 2011; Sánchez et al, 2018; Yoplac et al, 2019). However, the authors did not find any evidence about its use in the prediction of mechanical properties in whey protein‐based films.…”