1College of Food Science and Engineering, Yangzhou University, Yangzhou, China;
2School of Tourism and Cuisine, Yangzhou University, Yangzhou, China
The feasibility of near-infrared (NIR) spectroscopy and partial least-squares regression (PLSR) was investigated for rapid prediction of the quality of Chinese soy sauce. Twenty-four soy sauce samples from eight common brands available in China were analyzed for the contents of various components that may affect the quality of soy sauce. Sensory evaluation was also conducted to determine the relationship between components and the sensory quality of soy sauce. Subsequently, NIR spectra (400–2500 nm) of the samples were obtained, and the raw spectra were subjected to different pretreatment methods. PLSR was performed on the raw and treated spectra to construct models using a calibration set. The performance of models was evaluated by comparing the determination coefficient of prediction (R2P) and root-mean-square error of prediction (RMSEP). The results showed that the models constructed using the moisture content (R2P of 0.825 and RMSEP of 1.73), amino acid nitrogen content (R2P of 0.785 and RMSEP of 0.071), and taste scores (R2P of 0.733 and RMSEP of 11.93) performed well, and the interactions between amino acid nitrogen content and taste of soy sauce were clarified. This study demonstrates that NIR spectroscopy can be used as a valid alternative method for rapid prediction of the sensory quality of soy sauce during processing.
Key words: near-infrared (NIR) spectroscopy, partial least-squares regression (PLSR), rapid quality prediction, sensory evaluation, soy sauce
*Corresponding Author: Shuo Wang, College of Food Science and Engineering, Yangzhou University, Yangzhou, Jiangsu 225127, China. Email: [email protected]
Received: 20 August 2022; Accepted: 19 December 2022; Published: 17 January 2023
© 2023 Codon Publications
This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0). License (http://creativecommons.org/licenses/by-nc-sa/4.0/)
Soy sauce is a condiment originating in China that is made by fermentation of soybeans. It is widely used in the East and Southeast Asian cuisines. China is a major producer and consumer of soy sauce. The current global annual production volume of soy sauce is approximately 8 million tons, and China produces approximately 5 -million tons of this product (Shurtleff and Aoyagi, 2012). Compared with other countries, China has a wider variety of types of soy sauce, including braised soy sauce, steamed soy sauce, seafood soy sauce, and shrimp seed soy sauce. With improvements in quality of life, consumers are increasingly focusing on the sensory qualities of food (Jürkenbeck and Spiller, 2021). Consequently, the sensory characteristics of soy sauce have become important for determining its quality. Sensory evaluation is an important tool that can be used to explore consumer preferences and market demand (Stone et al., 2020).
Sensory evaluation of food products involves the use of human sensory organs to assess their various quality characteristics and provides comparative descriptions (Sarkar et al., 2022b; Steinhaus and Schieberle, 2007). However, this process is time-consuming and costly, and the results are highly subjective. Sensory evaluation can also be influenced by environmental factors (Xu et al. 2013), and it cannot be used for rapid testing of product sensory quality (Sarkar et al., 2022a). Therefore, establishment of an objective, effective, and rapid quality evaluation model is currently considered to be the most critical aspect in quality testing of soy sauce.
Near-infrared (NIR) spectroscopy is a convenient, efficient, and low-cost analytical technique that has undergone rapid development in the recent years (Zareef et al., 2020). It combines spectroscopic measurement analysis with chemometrics and is rapid, nonpolluting, nondestructive, and capable of determining multiple components simultaneously (Bázár et al., 2016; Li et al., 2007, 2020). Currently, NIR spectroscopy is widely used in food testing (Cortes et al., 2017; Escribano et al., 2017; Jamshidi et al., 2016; Lan et al., 2020; Lorenzo et al., 2009), such as identification of chestnut varieties (Corona et al., 2021), prediction of the moisture content in roasted pistachio kernels (Mohammadi-Moghaddam et al., 2018), and evaluation of the sensory properties of wines (Cayuela et al., 2017). Wang et al. (2021) investigated the feasibility of using NIR spectroscopy and partial least-squares regression (PLSR) to evaluate the quality of Japanese fermented soy sauce. They found that NIR spectroscopy could be used as an alternative to conventional methods for soy sauce quality control, and could quickly and economically grade soy sauce products. However, the brewing of Japanese soy sauce is different from that of Chinese soy sauce, and there is only one main variety (Koikuchi soy sauce) that differs significantly from Chinese soy sauce in its taste and texture (Diez-Simon et al., 2020). Consequently, the feasibility of applying NIR spectroscopy to sensory evaluation of Chinese soy sauce is unclear.
The aim of this study was to explore the feasibility of using NIR spectroscopy for rapid and objective sensory evaluation of Chinese soy sauce. The relationship between the sensory characteristics of Chinese soy sauce and their quality was first investigated using 24 samples from common soy sauce brands. The results were believed to provide scientific data for identifying key components of soy sauce and to improve their appearance and taste. Subsequently, the correlation between the components, sensory scores, and NIR spectrum were clarified by performing different pretreatment methods on the spectral data. Overall, NIR spectroscopy was promising to rapidly assess the sensory evaluation of Chinese soy sauce.
Twenty-four bottles of eight commercially available soy sauce brands were obtained. The bottles were randomly grouped into a calibration set (n = 18) and a validation set (n = 6). The calibration set was used to construct a quantitative calibration model, and the validation set was used to predict the model accuracy. The soy sauce samples were divided into two groups, A and B, according to their sensory scores. The 12 samples with higher sensory scores were placed in Group A, and the 12 samples with lower sensory scores were placed in Group B. The groups were used to compare the characteristics between different soy sauces.
The color, salt content, Brix value, pH, moisture content, and amino acid nitrogen content were analyzed in triplicates using the previously described method (Wang et al., 2018). Briefly, the color of each soy sauce sample was measured using a spectrophotometer (CM-5, Konica Minolta, Tokyo, Japan); the analyses of salt content and Brix values were carried out by using a conductivity salinometer (PAL-SALT, Atago Corporation, Tokyo, Japan) and a digital saccharimeter (PAL-1, Atago Co), respectively; the pH of the soy sauce was measured using a pH meter (Remag PHS-2F, Yidian Scientific Instruments Corporation, Shanghai, China); and the measurement of moisture content was performed by thermophysical drying. The determination of the amino acid nitrogen content in soy sauce was carried out by absorbance measurements at 400 nm, according to the national standards method reported in GB 5009.235-2016 Determination of amino acid nitrogen in food.
The contents of organic acids and sugars were determined by using a high-performance liquid chromatography (HPLC) as previously reported (Sarkar et al., 2020). The HPLC measurement was carried out by using the Shimadzu LC-20A system (Shimadzu Corporation, Kyoto, Japan). For the separation of organic acid, a Shodex KC-811 (i.d. 8 mm × 300 mm, Showa Denko Corporation, Tokyo, Japan) column was used, and the temperature was maintained at 50°C. The eluent (3.0 mmol/L HClO4 solution) flow rate was 1.0 mL/min, and the chromatogram of -samples was recorded with UV detection at 210 nm. As for sugar analysis, a Shodex KS-801 column (i.d. 8 mm × 300 mm, Showa Denko Co.) was used for separation, the column temperature was set at 80°C, ultrapure water used as eluent with a flow rate of 0.7 mL/min, and the detection was performed on a differential refractive index detector. Both analyses were measured in triplicates.
The sensory evaluation of soy sauce was carried out by employing 37 assessors (19–30 years old; 12 men and 25 women) with professional training and experience. For evaluation process, the 24 soy sauce samples were first poured onto white plastic plates to observe the color and texture. The assessors also put their noses 5 cm above the soy sauce to evaluate the aroma of soy sauce. Next, fresh cucumbers were cut into small pieces and dipped in the soy sauce for taste tests. Finally, each sample was scored according to its appearance, texture, taste, and aroma on a scale of 1–5 for each component. A higher score indicated higher satisfaction. The best possible score of a sample was 20 for each evaluation. The ratings given by all 37 individuals were added together to obtain a final total sensory evaluation score for each sample.
The acquisition of NIR spectroscopy for assessing the quality of soy sauces was performed according to a previous method (Wang et al., 2021). Briefly, the NIR spectra were collected using diffuse reflectance mode (Cary 5000, Varian Corporation, California, USA) with the wavelength ranging from 400 to 2500 nm. All samples were scanned on the same day to ensure that all measurement conditions were consistent, including ambient temperature and humidity. The pretreatment methods of spectra could be used to eliminate errors caused by disturbances, such as high--frequency random noise, baseline drift, and stray light, and to improve the reliability of the NIR model (Chen et al., 2013). In this work, nine pretreatments including first derivative (1st derivative), second derivative (2nd derivative), multiple scattering correction (MSC), and standard normal variate (SNV) methods, and combinations of these methods were employed into the spectrum processing, and the Savitsky–Golay algorithm with 10 points of smoothing was used to optimize the raw spectroscopic data.
The NIR predictive model was established by correlating the spectra and measured data with the aim to predicting the values of unknown samples (Xie et al., 2009). In this study, all soy sauce samples were randomly separated into two subsets of 16 and 8 samples. The 16 samples were classified into the calibration set and used for model development and cross validation, while the other 8 samples were classified into the validation set and used to test the practical performance of the established models.
The measured indicators, sensory scores, and spectral data of calibration set were first used to establish the predictive model. The modelling process was performed by PLSR using the Unscrambler data processing software (version 10.4, CAMO Software, Oslo, Norway). Subsequently, the established calibration models were validated by both internal full cross-validation and external validation of the validation set (Cámara-Martos et al., 2012). The correction coefficient of determination (R2c), root-mean-square error of correction (RMSEC), cross-validation coefficient of determination (R2cv), and root-mean-square error of cross-validation (RMSECV) were calculated to clarify the performance. The validation set was further employed to evaluate the feasibility of calibration model according to prediction coefficient of determination (R2p), root-mean-square error of prediction (RMSEP), and deviation rate (bias). For these indexes, R2 indicates the degree of linear correlation between the predicted value from the model and the reference value. While RMSEC, RMSECV, and RMSEP represent the standard deviations between model predictions and reference values during model calibration, cross-validation, and independent validation, respectively. For the same batch of samples, the smaller values of RMSEC, RMSECV, and RMSEP indicate better model performance. The calculation of these indexes was as following:
Where, m is the number of samples in the model.
Where, n is the number of cycles of cross-validation.
Where, k is the number of samples in the validation set used for model testing.
Where, k is the number of samples in the prediction set.
Among these equations, yi,actual denotes the measured value of the i-th sample, ŷi,actual denotes the average values in the calibration set (or validation set), and yi,predicted denotes the predicted value of the i-th sample.
The means and standard deviation were calculated to analyze the sensory attributes in soy sauce samples tested. All the results were reported as mean ± standard deviation of at least three measurements. The analysis of variance (ANOVA) was applied by using DPS software (Data Processing System, Hangzhou Rui feng Information Technology Co., Ltd., Zhejiang, China) to determine significant differences between soy sauces. The statistical significance level was set at P < 0.05.
In the sensory evaluation, the highest score was 512 and the lowest score was 332 (Table 1). Samples with high taste and appearance scores tended to have high overall scores. According to the results of the sensory evaluation, the 24 samples were divided into two groups of 12 with the higher-ranking samples in Group A and the -lower-ranking samples in Group B.
Table 1. Reference data on the sensory scores of soy sauce.
The results of color measurements are shown in Figure 1. L*, a*, b* are the three elements of the lab color space. L* indicates the brightness, with an L* value of 0 indicating pure black and an L* value of 100 indicating pure white. a* denotes the red index and indicates a spectral change from red to green. Larger positive values of a* indicate a reddish color and smaller negative values indicate a greenish color. The value of b* indicates a spectral change from yellow to blue, with larger positive values indicating a yellowish color and smaller negative values indicating a bluish color (Aliakbarian et al., 2016).
Figure 1. Measurements of a* and b* values of samples by using a CM-5 spectrophotometer. The numbering in figure is the same as the ranking in Table 1.
The a* values of soy sauce samples with higher rankings ranged from 0.2 to 1.0, and the b* values ranged from -0.1 to 0.4 (Figure 1). These results were within an appropriate range for the desired appearance of soy sauce. It is generally believed that as the Japanese soy sauce becomes redder (i.e., a larger a* value), the quality improves (Wang et al., 2021). However, the results in this study showed that the a* value of Chinese soy sauce could vary within a certain range. Very high values of a* and b* will not improve the quality of soy sauce. The L* values of the soy sauce ranged from 2.6 to 3.2 (Table 2), and did not greatly contribute to its appearance.
Table 2. Average physicochemical compositions in soy sauce samples.
|Rank||a*||b*||L*||Salt (%)||Brix (%)|
|1||0.55 ± 0.06||−0.03 ± 0.02||2.64 ± 0.01||12.78 ± 0.15||46.66 ± 0.55|
|2||0.31 ± 0.02||0.25 ± 0.06||3.12 ± 0.04||15.57 ± 0.62||41.78 ± 1.61|
|3||0.51 ± 0.02||0.01 ± 0.03||2.84 ± 0.02||13.40 ± 0.06||45.79 ± 1.16|
|4||0.86 ± 0.04||0.36 ± 0.04||2.86 ± 0.01||15.45 ± 0.75||51.18 ± 0.97|
|5||0.23 ± 0.06||0.15 ± 0.02||2.71 ± 0.04||14.42 ± 0.17||43.29 ± 0.55|
|6||0.80 ± 0.03||0.33 ± 0.02||2.92 ± 0.02||13.97 ± 0.15||52.42 ± 0.57|
|7||0.14 ± 0.05||0.04 ± 0.04||2.83 ± 0.09||17.89 ± 0.06||41.45 ± 1.12|
|8||1.00 ± 0.06||0.07 ± 0.05||3.07 ± 0.01||15.78 ± 0.34||42.45 ± 1.45|
|9||0.96 ± 0.04||0.28 ± 0.04||3.54 ± 0.03||12.96 ± 0.32||41.33 ± 1.52|
|10||2.08 ± 0.03||1.24 ± 0.04||3.28 ± 0.01||14.45 ± 0.46||45.38 ± 0.97|
|11||0.70 ± 0.04||0.27 ± 0.01||2.99 ± 0.02||10.56 ± 0.16||32.86 ± 1.21|
|12||1.11 ± 0.03||0.43 ± 0.03||3.19 ± 0.02||13.46 ± 0.30||41.19 ± 0.54|
|13||2.06 ± 0.07||0.99 ± 0.05||3.69 ± 0.02||7.18 ± 0.15||37.66 ± 0.55|
|14||1.83 ± 0.03||1.06 ± 0.07||3.38 ± 0.01||11.90 ± 0.30||35.56 ± 1.15|
|15||1.40 ± 0.07||0.33 ± 0.01||3.42 ± 0.04||16.45 ± 0.86||40.41 ± 1.09|
|16||1.40 ± 0.02||0.63 ± 0.02||3.08 ± 0.01||15.56 ± 0.19||41.87 ± 0.53|
|17||1.48 ± 0.05||0.63 ± 0.02||3.03 ± 0.01||14.31 ± 0.11||52.31 ± 1.62|
|18||1.84 ± 0.03||0.83 ± 0.04||3.45 ± 0.06||13.24 ± 0.28||46.37 ± 0.53|
|19||2.09 ± 0.01||1.51 ± 0.02||4.62 ± 0.04||16.38 ± 0.00||32.36 ± 0.98|
|20||0.24 ± 0.02||−0.19 ± 0.03||1.34 ± 0.03||13.77 ± 0.26||56.24 ± 1.77|
|21||0.27 ± 0.07||−0.19 ± 0.05||2.19 ± 0.01||14.92 ± 0.72||54.19 ± 0.57|
|22||2.47 ± 0.03||1.11 ± 0.04||3.61 ± 0.01||15.55 ± 0.31||50.44 ± 0.56|
|23||−0.02 ± 0.02||−0.28 ± 0.04||2.12 ± 0.02||15.43 ± 0.21||40.63 ± 1.72|
|24||0.32 ± 0.04||0.13 ± 0.05||2.50 ± 0.01||16.27 ± 0.49||49.21 ± 0.01|
|Rank||Moisture (%)||Glucose*||Galactose||Oxalic acid||Citric acid|
|1||59.11 ± 0.05||12.95 ± 0.56||ND||1.61 ± 0.05||5.12 ± 0.01|
|2||64.07 ± 0.03||5.76 ± 0.09||ND||0.54 ± 0.02||6.42 ± 0.15|
|3||63.52 ± 0.01||7.48 ± 0.05||ND||0.82 ± 0.05||17.21 ± 0.19|
|4||60.84 ± 0.01||11.01 ± 0.10||6.07 ± 0.15||0.59 ± 0.01||8.57 ± 0.03|
|5||61.46 ± 0.11||7.66 ± 0.24||ND||0.76 ± 0.04||11.59 ± 0.12|
|6||59.42 ± 0.04||6.54 ± 0.06||7.96 ± 0.29||0.61 ± 0.00||14.06 ± 0.18|
|7||60.52 ± 0.58||8.62 ± 0.08||ND||0.65 ± 0.02||9.74 ± 0.39|
|8||64.89 ± 0.09||3.25 ± 0.02||7.94 ± 0.43||1.69 ± 0.00||5.11 ± 0.19|
|9||63.31 ± 0.10||11.94 ± 0.21||6.85 ± 0.14||0.91 ± 0.04||15.15 ± 0.47|
|10||63.15 ± 0.08||10.85 ± 0.30||8.10 ± 0.11||0.94 ± 0.02||7.83 ± 0.16|
|11||72.76 ± 0.13||5.68 ± 0.16||4.34 ± 0.18||0.66 ± 0.02||5.29 ± 0.18|
|12||63.87 ± 0.57||12.51 ± 0.12||10.32 ± 0.14||2.44 ± 0.05||7.10 ± 0.33|
|13||66.52 ± 0.29||3.25 ± 0.03||4.46 ± 0.07||0.29 ± 0.01||6.33 ± 0.16|
|14||61.69 ± 0.11||13.22 ± 0.35||9.85 ± 0.37||1.05 ± 0.02||8.66 ± 0.45|
|15||67.33 ± 0.20||5.29 ± 0.05||ND||0.86 ± 0.01||6.40 ± 0.16|
|16||63.73 ± 0.10||10.86 ± 0.12||6.80 ± 0.37||0.91 ± 0.01||5.41 ± 0.13|
|17||60.82 ± 0.34||9.65 ± 0.36||8.35 ± 0.03||0.60 ± 0.00||14.03 ± 0.07|
|18||60.24 ± 0.10||10.55 ± 0.05||ND||0.97 ± 0.03||7.02 ± 0.12|
|19||71.03 ± 0.30||7.57 ± 0.10||3.75 ± 0.16||0.58 ± 0.02||5.05 ± 0.14|
|20||47.25 ± 0.39||18.57 ± 0.13||ND||30.08 ± 0.93||ND|
|21||51.30 ± 0.13||11.62 ± 0.13||ND||10.76 ± 0.92||6.83 ± 0.21|
|22||64.21 ± 0.01||11.28 ± 0.14||8.17 ± 0.14||0.85 ± 0.04||5.61 ± 0.43|
|23||61.32 ± 0.64||21.95 ± 0.34||ND||8.39 ± 0.14||ND|
|24||62.98 ± 0.11||5.81 ± 0.10||ND||1.82 ± 0.03||7.57 ± 0.32|
|Rank||Tartaric acid||Lactic acid||Pyroglutamic acid||Amino acid nitrogen|
|1||35.96 ± 0.97||3.97 ± 0.11||2.64 ± 0.05||1.06 ± 0.00|
|2||60.56 ± 1.95||16.87 ± 0.16||4.33 ± 0.05||0.93 ± 0.00|
|3||82.52 ± 1.19||12.05 ± 0.39||1.76 ± 0.08||1.01 ± 0.00|
|4||10.12 ± 0.04||ND||4.19 ± 0.14||1.05 ± 0.00|
|5||27.11 ± 0.37||ND||5.19 ± 0.03||1.08 ± 0.00|
|6||9.86 ± 0.13||ND||4.43 ± 0.05||1.05 ± 0.00|
|7||98.03 ± 3.60||ND||5.04 ± 0.07||1.01 ± 0.00|
|8||3.01 ± 0.12||9.66 ± 0.41||2.18 ± 0.06||0.88 ± 0.00|
|9||54.18 ± 2.31||ND||5.47 ± 0.20||1.07 ± 0.00|
|10||20.99 ± 0.59||ND||4.19 ± 0.14||0.98 ± 0.00|
|11||8.68 ± 0.36||17.78 ± 0.38||3.89 ± 0.08||0.72 ± 0.00|
|12||12.40 ± 0.28||22.73 ± 0.78||4.67 ± 0.10||0.90 ± 0.00|
|13||3.78 ± 0.15||8.67 ± 0.29||2.16 ± 0.04||0.85 ± 0.00|
|14||25.48 ± 0.92||ND||5.16 ± 0.19||0.91 ± 0.00|
|15||5.20 ± 0.13||6.09 ± 0.04||2.94 ± 0.06||0.79 ± 0.00|
|16||11.96 ± 0.34||ND||4.70 ± 0.12||1.10 ± 0.00|
|17||12.27 ± 0.05||ND||4.81 ± 0.20||0.93 ± 0.00|
|18||14.89 ± 0.36||15.21 ± 0.20||2.90 ± 0.04||0.52 ± 0.00|
|19||9.82 ± 0.33||4.18 ± 0.15||1.33 ± 0.05||0.48 ± 0.00|
|20||11.71 ± 0.30||ND||2.58 ± 0.02||1.00 ± 0.00|
|21||54.27 ± 2.09||ND||4.30 ± 0.13||0.98 ± 0.00|
|22||14.47 ± 0.59||9.42 ± 0.14||4.73 ± 0.13||1.09 ± 0.00|
|23||9.46 ± 0.46||ND||3.88 ± 0.09||0.66 ± 0.00|
|24||20.37 ± 0.89||ND||4.04 ± 0.09||0.78 ± 0.00|
*The units of organic acids, sugars, and amino acid nitrogen are g/kg; ND: not detected.
A certain amount of brine is usually added to soy sauce during the fermentation process to inhibit the growth of unwanted bacteria (Syifaa et al., 2016). The salt contents and Brix values of the 24 soy sauce samples are shown in Figure 2. The high-quality soy sauce samples had salt contents between 12 and 17% and Brix values between 41 and 47%. If the salt content is too high, it will adversely affect the taste and health benefits of the soy sauce (Bibbins-Domingo et al., 2010). By contrast, if the salt content is too low, the growth of unwanted bacteria during the fermentation process will not be inhibited and this affects the quality of the soy sauce (Taormina et al., 2010).
Figure 2. Plots of salt content versus Brix content for soy sauces. The numbering in figure is the same as the ranking in Table 1.
The moisture content affects both the fermentation process of the soy sauce and the texture of the finished product. The moisture contents of the soy sauce samples are shown in Table 2. The results showed that moisture contents did not adversely affect the sensory quality of the soy sauce.
The pH values and amino acid nitrogen contents of the samples are shown in Figure 3. The pH of soy sauce affects its flavor, with a lower pH resulting in a more prominent sour taste. The pH values of the higher--quality soy sauces were all greater than 5.1, and the pH values of the lower-quality soy sauces were all less than 4.9. These results showed that higher acidity might be detrimental to the taste of soy sauce. Amino acid nitrogen represents the nitrogen content of free amino acids in soy sauce, and amino acids are closely related to the umami taste of soy sauce (Yanfang and Wenyi, 2009). The average levels of amino acid nitrogen in Groups A and B were 0.978 and 0.841 g/kg, respectively. The difference in the amino acid nitrogen between Groups A and B was significant (P < 0.05), and this might be related to the taste results. Consequently, the samples with high sensory scores had high amino acid nitrogen contents, which supported the positive effect of the amino acid nitrogen content on the taste of soy sauce.
Figure 3. Plots of pH and amino acid nitrogen content for soy sauces. The numbering in figure is the same as the ranking in Table 1.
Five organic acids including oxalic, citric, tartaric, lactic, and pyroglutamic acids were detected in this experiment (Table 2). Among these acids, tartaric acid had the highest content. The average contents of oxalic, citric, tartaric, lactic, and pyroglutamic acids were 1.018, 9.433, 35.285, 6.922, and 3.998 g/kg, respectively, in Group A, and 4.763, 6.076, 16.14, 3.631, and 3.628 g/kg, respectively, in Group B. These results showed that amino acid nitrogen, citric acid, tartaric acid, and lactic acid are beneficial to the taste of soy sauce, while oxalic acid is detrimental to the taste of soy sauce and pyroglutamic acid has little effect on the taste of soy sauce.
The sugars in soy sauce provide sweetness and act as a source of carbon for other chemical reactions during fermentation (Chiou et al., 1999; Kwak and Lim, 2004). Both glucose and galactose were detected in the -samples (Table 2), with glucose present at higher -levels. Glucose was detected in all of the samples at 3.25–21.95 g/kg. The average level in Group A (8.688 g/kg) was lower than that in Group B (10.802 g/kg). The lower level in Group A might be because glucose is an important carbon source that is consumed during the later stages of fermentation to produce some taste and flavor substances. This reduces the level of glucose and makes the overall flavor of the soy sauce richer. Galactose was detected in only 13 of the samples (Table 2) at 3.75–10.32 g/kg, and there was no significant correlation between the taste quality of the soy sauce and galactose. Overall, the physical and chemical properties could be used to objectively evaluate the quality of soy sauce.
In the NIR spectra (Figure 4), the samples showed multiple absorption peaks. Large differences were observed in the 400–800 nm region, which was probably because of the different colors of the soy sauce samples. All spectra showed large absorption peaks at 996–1134 nm, 1134–1325 nm, 1800–1950 nm, and 2140–2380 nm. The first two absorption peaks corresponded to the C-O and O-H functional groups in alcohols and phenols, and the latter two were attributed to O-H, N-H, and C-H groups.
Figure 4. Original near-infrared spectra of soy sauces tested.
PLSR was used to build the model, and nine preprocessing methods (no treatment, 1st derivative, 2nd derivative, MSC, MSC + 1st derivative, MSC + 2nd derivative, SNV, SNV + 1st derivative, SNV + 2nd derivative) were used to improve the model accuracy. The data obtained by modelling and prediction using PLSR are shown in Table S1 Accuracy parameters for judging the quality of the PLSR models based on the calibration and validation sets of different components and sensory scores of soy sauce samples. When the NIR spectra were modelled using the total sensory evaluation score, the model had poor predictive performance as shown by the low R2P and high RMSEP (Table 3). The four scores were then modelled separately, and the model with the taste score was found to have the best predictive performance. The best taste score model was obtained using MSC + 1st derivative, which had a R2P of 0.733 and RMSEP of 11.93 (Table 3). The appearance score model was similar to the total sensory evaluation score model and had poor predictive performance. The aroma score model and the texture score model gave no valid prediction set data, which indicated that they had poor feasibility. Each parameter that may affect the quality of soy sauce was modelled with the spectra, and the moisture and amino acid nitrogen contents had the best modelling results and predictive ability. The best moisture content model was developed using MSC and had a R2P of 0.825 and RMSEP of 1.73. The best amino acid nitrogen content model was developed using SNV correction and had a R2P of 0.785 and RMSEP of 0.071. Both models had good predictive performance. The predictive performances of the models constructed using the remaining constituents were poor because either there was no valid prediction set data or the models had low R2P and high RMSEP. In summary, the models for the moisture content, amino acid nitrogen content, and taste score had good predictive performances and could be used to rapidly predict the soy sauce quality.
Table 3. Accuracy parameters for judging the quality of the PLSR models based on the calibration and validation sets of sensory scores of soy sauce samples using the full spectrum.
|Total sensory evaluation score||None||3||0.313||30.122||0.152||35.434||0.222||59.621||−1.310|
|MSC +1st derivative||2||0.345||29.400||0.236||33.623||0.218||59.823||2.104|
|MSC +2nd derivative||3||0.434||27.333||0.196||34.498||0.234||59.197||1.189|
|SNV +1st derivative||2||0.348||29.347||0.241||33.518||0.227||59.465||2.108|
|SNV +2nd derivative||2||0.440||27.202||0.208||34.241||0.236||59.106||1.357|
|MSC +1st derivative||1||0.564||8.560||0.524||9.476||0.733||11.930||1.486|
|MSC +2nd derivative||1||0.563||8.572||0.509||9.615||0.727||12.058||1.016|
|SNV +1st derivative||1||0.658||7.584||0.539||9.318||0.687||12.930||1.671|
|SNV +2nd derivative||1||0.555||8.646||0.507||9.642||0.730||11.991||1.149|
|Amino acid nitrogen||None||5||0.78||0.081||0.51||0.129||NA||0.177||0.085|
|MSC +1st derivative||2||0.75||0.087||0.610||0.114||NA||0.186||0.084|
|MSC +2nd derivative||3||0.98||0.023||0.627||0.1116||0.451||0.114||0.049|
|SNV +1st derivative||2||0.74||0.087||0.607||0.1146||NA||0.186||0.083|
|SNV +2nd derivative||3||0.99||0.021||0.626||0.1117||0.33||0.126||0.052|
|MSC +1st derivative||1||0.955||1.132||0.831||2.312||0.703||2.261||−0.169|
|MSC +2nd derivative||1||0.714||2.838||0.697||3.097||NA||7.428||−3.822|
|SNV +1st derivative||1||0.956||1.119||0.793||2.559||0.689||2.314||−0.269|
|SNV +2nd derivative||1||0.709||2.864||0.686||3.149||NA||7.286||−3.765|
LVs, number of latent variables; MSC, multiple scattering correction; NA, not available; R2c, the correction coefficient of determination; R2cv, the coefficient of determination in full cross-validation; R2p, the coefficient of determination for prediction; RMSEC, root mean square error of correction; RMSECV, root mean square error of cross-validation; RMSEP, root mean square error of prediction; SNV: standard normal variate.
Amino acid nitrogen is the main umami substance in soy sauce. The taste analysis showed that soy sauce samples with high sensory scores and good taste also had high amino acid nitrogen contents. The good predictive ability of NIR spectroscopy for both the amino acid nitrogen content and taste score was likely related to the close relationship between these two components. Generally, high-quality soy sauce contains more amino acid nitrogen than low-quality soy sauce. This means that amino acid nitrogen is important for evaluating the quality of soy sauce.
The rich substances in soy sauce that are related to sensory quality have brought new challenges on how to quantitatively evaluate its final quality. This study performed a comprehensive investigation of physicochemical and sensory experiments on 24 commercial soy sauce samples in China. The contents of amino acid nitrogen, citric acid, tartaric acid, and lactic acid, beneficial for improving the quality of soy sauce, and the interaction between them were clarified. Glucose showed a negative effect on the soy sauce quality. Furthermore, with the aim to developing a rapid and objective method for predicting the measured components and sensory scores of soy sauces, the feasibility of NIR spectroscopy was tested and validated by modelling with PLSR. The predictive models showed good performances on predicting the moisture content, amino acid nitrogen content, and the taste score of soy sauce, which were considered useful for classifying the sensory quality of soy sauce quickly and economically. Finally, this work has some limitations, such as ignoring the aroma-producing compounds. Therefore, further studies should focus on analyzing the factors that may affect the quality of soy sauce, and improve the predictive performance for routine use.
Authors have no conflicts of interest to declare for this article.
Aliakbarian, B., Bagnasco, L., Perego, P., Leardi, R. and Casale, M., 2016. UV-VIS spectroscopy for monitoring yogurt stability during storage time. Analytical Methods 8(30): 5962–5969. 10.1039/C6AY00607H
Bázár, G., Romvári, R., Szabó, A., Somogyi, T., Éles, V. and Tsenkova, R., 2016. NIR detection of honey adulteration reveals differences in water spectral pattern. Food Chemistry 194: 873–880. 10.1016/j.foodchem.2015.08.092
Bibbins-Domingo, K., Chertow, G.M., Coxson, P.G., Moran, A., Lightwood, J.M., Pletcher, M.J., et al., 2010. Projected effect of dietary salt reductions on future cardiovascular disease. New England Journal of Medicine 362(7): 590–599. 10.1056/NEJMoa0907355
Cámara-Martos, F., Zurera-Cosano, G., Moreno-Rojas, R., García-Gimeno, R.M. and Pérez-Rodríguez, F., 2012. Identification and quantification of lactic acid bacteria in a water-based matrix with near-infrared spectroscopy and multivariate regression modeling. Food Analytical Methods 5(1): 19–28. 10.1007/s12161-011-9221-5
Cayuela, J.A., Puertas, B. and Cantos-Villar, E., 2017. Assessing wine sensory attributes using Vis/NIR. European Food Research and Technology 243(6): 941–953. 10.1007/s00217-016-2807-9
Chen, H., Song, Q., Tang, G., Feng, Q. and Liang, L., 2013. The combined optimization of savitzky-golay smoothing and multiplicative scatter correction for FT-NIR PLS models. ISRN Spectroscopy 2013: 1–9. 10.1155/2013/642190
Chiou, R.Y., Ferng, S. and Beuchat, L.R., 1999. Fermentation of low-salt miso as affected by supplementation with ethanol. International Journal of Food Microbiology 48(1): 11–20. 10.1016/S0168-1605(99)00033-1
Corona, P., Frangipane, M.T., Moscetti, R., Lo Feudo, G., Castellotti, T. and Massantini, R., 2021. Chestnut cultivar identification through the data fusion of sensory quality and FT-NIR spectral data. Foods 10(11): 2575. 10.3390/foods10112575
Cortes, V., Rodriguez, A., Blasco, J., Rey, B., Besada, C., Cubero, S., et al., 2017. Prediction of the level of astringency in persimmon using visible and near-infrared spectroscopy. Journal of Food Engineering 204: 27–37. 10.1016/j.jfoodeng.2017.02.017
Diez-Simon, C., Eichelsheim, C., Mumm, R. and Hall, R.D., 2020. Chemical and sensory characteristics of soy sauce: a review. Journal of Agricultural and Food Chemistry 68(42): 11612–11630. 10.1021/acs.jafc.0c04274
Escribano, S., Biasi, W.V., Lerud, R., Slaughter, D.C. and Mitcham, E.J., 2017. Non-destructive prediction of soluble solids and dry matter content using NIR spectroscopy and its relationship with sensory quality in sweet cherries. Postharvest Biology and Technology 128: 112–120. 10.1016/j.postharvbio.2017.01.016
Jamshidi, B., Mohajerani, E. and Jamshidi, J., 2016. Developing a Vis/NIR spectroscopic system for fast and non-destructive pesticide residue monitoring in agricultural product. Measurement 89: 1–6. 10.1016/j.measurement.2016.03.069
Jürkenbeck, K. and Spiller, A., 2021. Importance of sensory quality signals in consumers’ food choice. Food Quality and Preference 90: 104155. 10.1016/j.foodqual.2020.104155
Kwak, E.J. and Lim, S.I., 2004. The effect of sugar, amino acid, metal ion, and NaCl on model Maillard reaction under pH control. Amino Acids 27(1): 85–90. 10.1007/s00726-004-0067-7
Lan, W., Jaillais, B., Leca, A., Renard, C.M. and Bureau, S., 2020. A new application of NIR spectroscopy to describe and predict purees quality from the non-destructive apple measurements. Food Chemistry 310: 125944. 10.1016/j.foodchem.2019.125944
Li, X., He, Y. and Fang, H., 2007. Non-destructive discrimination of Chinese bayberry varieties using Vis/NIR spectroscopy. Journal of Food Engineering 81(2): 357–363. 10.1016/j.jfoodeng.2006.10.033
Li, X., Zhang, L., Zhang, Y., Wang, D., Wang, X., Yu, L., et al., 2020. Review of NIR spectroscopy methods for nondestructive quality analysis of oilseeds and edible oils. Trends in Food Science & Technology 101: 172–181. 10.1016/j.tifs.2020.05.002
Lorenzo, C., Garde-Cerdán, T., Pedroza, M.A., Alonso, G.L. and Salinas, M.R., 2009. Determination of fermentative volatile compounds in aged red wines by near infrared spectroscopy. Food Research International 42(9): 1281–1286. 10.1016/j.foodres.2009.03.021
Mohammadi-Moghaddam, T., Razavi, S., Sazgarnia, A. and Taghizadeh, M., 2018. Predicting the moisture content and textural characteristics of roasted pistachio kernels using Vis/NIR reflectance spectroscopy and PLSR analysis. Journal of Food Measurement and Characterization 12(1): 346–355. 10.1007/s11694-017-9646-7
Sarkar, T., Mukherjee, A., Chatterjee, K., Shariati, M.A., Rebezov, M., Rodionova, S., et al., 2022a. Comparative analysis of statistical and supervised learning models for freshness assessment of oyster mushrooms. Food Analytical Methods 15(4): 917–939. 10.1007/s12161-021-02161-7
Sarkar, T., Salauddin, M., Hazra, S.K. and Chakraborty, R., 2020. The impact of raw and differently dried pineapple (Ananas comosus) fortification on the vitamins, organic acid and carotene profile of dairy rasgulla (sweetened cheese ball). Heliyon 6(10): e05233. 10.1016/j.heliyon.2020.e05233
Sarkar, T., Salauddin, M., Pati, S., Chakraborty, R., Shariati, M.A., Rebezov, M., et al., 2022b. The fuzzy cognitive map–based shelf-life modelling for food storage. Food Analytical Methods 15(3): 579–597. 10.1007/s12161-021-02147-5
Shurtleff, W. and Aoyagi, A., 2012. History of soy sauce (160 CE to 2012): extensively annotated bibliography and sourcebook. Soyinfo Center, Lafayette, CA, USA.
Steinhaus, P. and Schieberle, P., 2007. Characterization of the key aroma compounds in soy sauce using approaches of molecular sensory science. Journal of Agricultural and Food Chemistry 55(15): 6262–6269. 10.1021/jf0709092
Stone, H., Bleibaum, R. and Thomas, H.A., 2020. Sensory evaluation practices. 5th Edition, Elsevier Academic Press, Boston, MA, USA.
Syifaa, A.S., Jinap, S., Sanny, M. and Khatib, A., 2016. Chemical profiling of different types of soy sauce and the relationship with its sensory attributes. Journal of Food Quality 39(6): 714–725. 10.1111/jfq.12240
Taormina, P.J., 2010. Implications of salt and sodium reduction on microbial food safety. Critical Reviews in Food Science and Nutrition 50(3): 209–227. 10.1080/10408391003626207
Wang, S., Liu, X., Tamura, T., Kyouno, N., Zhang, H. and Chen, J.Y., 2021. Rapid detection of the quality of miso (Japanese fermented soybean paste) using visible/near-infrared spectroscopy. Analytical Letters 54(14): 2304–2314. 10.1080/00032719.2020.1858092
Wang, S., Zhang, H., Liu, X., Tamura, T., Kyouno, N. and Chen, J.Y., 2018. Relationship between chemical characteristics and sensory evaluation of koikuchi soy sauce. Analytical Letters 51(14): 1813–1827. 10.1080/00032719.2017.1419252
Xie, L., Ye, X., Liu, D. and Ying, Y., 2009. Quantification of glucose, fructose and sucrose in bayberry juice by NIR and PLS. Food Chemistry 114(3): 1135–1140. 10.1016/j.foodchem.2008.10.076
Xu, D., Li, C., Zhao, M., Fengc, Y., Sun, L. and Wang, Y., 2013. Assessment on the improvement of soy sauce fermentation by Aspergillus oryzae HG76. Biocatalysis and Agricultural Biotechnology 2(4): 344–351. 10.1016/j.bcab.2013.07.002
Yanfang, Z. and Wenyi, T., 2009. Biochemical changes in low-salt fermentation of solidstate soy sauce. African Journal of Biotechnology 8(24): 7028–7034
Zareef, M., Chen, Q., Hassan, M.M., Arslan, M., Hashim, M.M., Ahmad, W., et al., 2020. An overview on the applications of typical non-linear algorithms coupled with NIR spectroscopy in food analysis. Food Engineering Reviews 12(2): 173–190. 10.1007/s12393-020-09210-7
Table S1. Accuracy parameters for judging the quality of the PLSR models based on the calibration and validation sets of different components and sensory scores of soy sauce samples.
|Amino acid nitrogen||None||5||0.777||0.081||0.510||0.129||NA||0.177||0.085|
|MSC + 1st derivative||2||0.747||0.087||0.610||0.114||NA||0.186||0.084|
|MSC + 2nd derivative||3||0.982||0.023||0.627||0.116||0.451||0.114||0.049|
|SNV + 1st derivative||2||0.745||0.087||0.607||0.146||NA||0.186||0.083|
|SNV + 2nd derivative||3||0.985||0.021||0.626||0.117||0.330||0.126||0.052|
|MSC + 1st derivative||2||0.366||3.024||0.253||3.476||NA||5.012||3.826|
|MSC + 2nd derivative||2||0.271||3.243||0.155||3.697||NA||4.285||3.694|
|SNV + 1st derivative||2||0.367||3.022||0.254||3.475||NA||4.997||3.825|
|SNV + 2nd derivative||2||0.262||3.263||0.142||3.728||NA||4.253||3.708|
|MSC + 1st derivative||1||0.276||3.153||0.088||3.747||0.531||3.966||1.337|
|MSC + 2nd derivative||1||0.287||3.128||0.136||3.647||0.585||3.728||1.443|
|SNV + 1st derivative||1||0.271||3.163||0.077||3.769||0.513||4.039||1.310|
|SNV + 2nd derivative||1||0.282||3.140||0.125||3.669||0.568||3.806||1.414|
|MSC + 1st derivative||2||0.997||0.379||0.789||3.363||NA||9.785||3.831|
|MSC + 2nd derivative||3||0.959||1.404||0.604||4.611||NA||8.600||3.691|
|SNV + 1st derivative||2||0.997||0.384||0.774||3.482||NA||9.602||3.713|
|SNV + 2nd derivative||3||0.952||1.509||0.576||4.773||NA||8.328||3.594|
|MSC + 1st derivative||1||0.091||6.159||0.040||6.699||0.192||7.347||–1.330|
|MSC + 2nd derivative||1||0.074||6.215||0.047||6.676||0.156||7.506||–1.353|
|SNV + 1st derivative||1||0.090||6.161||0.036||6.712||0.196||7.326||–1.304|
|SNV + 2nd derivative||1||0.074||6.214||0.046||6.679||0.161||7.484||–1.335|
|MSC + 1st derivative||4||0.673||2.410||0.261||3.838||0.624||2.250||0.947|
|MSC + 2nd derivative||5||0.950||0.941||0.402||3.453||NA||4.073||0.200|
|SNV + 1st derivative||4||0.680||2.385||0.304||3.727||NA||5.088||0.487|
|SNV + 2nd derivative||5||0.950||0.943||0.406||3.441||NA||4.099||0.293|
|MSC + 1st derivative||1||0.955||1.132||0.831||2.312||0.703||2.261||–0.169|
|MSC + 2nd derivative||1||0.714||2.838||0.697||3.097||NA||7.428||–3.822|
|SNV + 1st derivative||1||0.956||1.119||0.793||2.559||0.689||2.314||–0.269|
|SNV + 2nd derivative||1||0.709||2.864||0.686||3.149||NA||7.286||–3.765|
|MSC + 1st derivative||1||0.443||4.849||0.397||5.343||NA||8.114||5.165|
|MSC + 2nd derivative||1||0.403||5.019||0.356||5.521||NA||8.658||5.294|
|SNV + 1st derivative||1||0.451||4.812||0.403||5.315||NA||7.993||5.129|
|SNV + 2nd derivative||1||0.412||4.982||0.368||5.470||NA||8.547||5.264|
|MSC + 1st derivative||1||0.684||0.369||0.614||0.432||NA||0.510||0.220|
|MSC + 2nd derivative||1||0.668||0.378||0.639||0.417||NA||0.537||0.197|
|SNV + 1st derivative||1||0.685||0.369||0.606||0.436||NA||0.505||0.227|
|SNV + 2nd derivative||1||0.671||0.376||0.637||0.419||NA||0.529||0.203|
|MSC + 1st derivative||1||0.392||0.386||0.346||0.424||NA||0.423||0.322|
|MSC + 2nd derivative||1||0.338||0.403||0.272||0.448||NA||0.414||0.316|
|SNV + 1st derivative||1||0.399||0.384||0.354||0.421||NA||0.426||0.324|
|SNV + 2nd derivative||1||0.347||0.400||0.286||0.443||NA||0.416||0.318|
|MSC + 1st derivative||1||0.433||0.497||0.403||0.540||NA||0.969||0.845|
|MSC + 2nd derivative||1||0.588||0.424||0.379||0.551||NA||0.929||0.833|
|SNV + 1st derivative||1||0.440||0.494||0.412||0.536||NA||0.972||0.849|
|SNV + 2nd derivative||1||0.592||0.422||0.378||0.551||NA||0.934||0.835|
|MSC + 1st derivative||2||0.374||10.821||0.294||12.168||0.334||10.256||–5.802|
|MSC + 2nd derivative||3||0.469||9.969||0.265||12.414||0.261||10.808||–6.382|
|SNV + 1st derivative||2||0.376||10.808||0.298||12.134||0.362||10.037||–5.790|
|SNV + 2nd derivative||3||0.475||9.914||0.277||12.314||0.300||10.520||–6.303|
|MSC + 1st derivative||1||0.564||8.560||0.524||9.476||0.733||11.930||1.486|
|MSC + 2nd derivative||1||0.563||8.572||0.509||9.615||0.727||12.058||1.016|
|SNV + 1st derivative||1||0.658||7.584||0.539||9.318||0.687||12.930||1.671|
|SNV + 2nd derivative||1||0.555||8.646||0.507||9.642||0.730||11.991||1.149|
|Total sensory evaluation score||None||3||0.313||30.122||0.152||35.434||0.222||59.621||–1.310|
|MSC + 1st derivative||2||0.345||29.400||0.236||33.623||0.218||59.823||2.104|
|MSC + 2nd derivative||3||0.434||27.333||0.196||34.498||0.234||59.197||1.189|
|SNV + 1st derivative||2||0.348||29.347||0.241||33.518||0.227||59.465||2.108|
|SNV + 2nd derivative||2||0.440||27.202||0.208||34.241||0.236||59.106||1.357|
|MSC + 1st derivative||2||0.303||9.703||0.163||11.259||NA||24.821||3.354|
|MSC + 2nd derivative||3||0.429||8.778||0.181||11.136||NA||24.610||3.231|
|SNV + 1st derivative||2||0.300||9.719||0.157||11.298||NA||24.824||3.296|
|SNV + 2nd derivative||3||0.429||8.778||0.149||11.351||NA||24.657||3.267|
LVs, number of latent variables; MSC, multiple scattering correction; NA, not available; R2c, the correction coefficient of determination; R2cv, the coefficient of determination in full cross-validation; R2p, the coefficient of determination for prediction; RMSEC, root mean square error of correction; RMSECV, root mean square error of cross-validation; RMSEP, root mean square error of prediction; SNV: standard normal variate.