Research Article

Received: 20 January 2012 Revised: 10 April 2012 Accepted: 16 April 2012 Published online in Wiley Online Library: 12 June 2012

(wileyonlinelibrary.com) DOI 10.1002/jsfa.5738

Determining the contents of protein and amino

acids in peanuts using near-infrared
reflectance spectroscopy
Li Wang, Qiang Wang,∗ Hongzhi Liu, Li Liu and Yin Du

Abstract
BACKGROUND: The protein and amino acid contents of peanuts play a key role in determining their quality and value.
Therefore, accurate, nondestructive, quick, and automated measurement of these components would be valuable in a
commercial environment. This study explored the feasibility of determining the contents of protein and amino acids in peanuts
using near infrared–reflectance spectroscopy (NIRS).

RESULTS: 141 peanut samples were collected from 12 provinces in China. The spectra were scanned and obtained with an NIRS
system. The determination coefficient and the ratio of the standard deviation in the validation set to the standard error of
validation corresponded to 0.99 and 6.53 for protein, 0.88 and 2.52 for Asp, 0.83 and 3.00 for Thr, 0.86 and 2.40 for Ser, 0.87
and 2.57 for Glu, 0.88 and 2.36 for Gly, 0.88 and 3.00 for Leu, 0.89 and 2.88 for Arg, and 0.96 and 7.50 for Cys.

CONCLUSIONS: NIRS combined with multivariate calibration has significant potential in determining the protein and amino
acid contents of peanuts. This method is suitable for use in an industrial setting owing to its ease of use as well as the relatively
low cost of obtaining and running the necessary equipment.

c 2012 Society of Chemical Industry

Keywords: nondestructive; peanuts; protein; amino acid; NIRS

INTRODUCTION acid, glutamic acid, phenylalanine, and histidine, react with

Peanut (Arachis hypogaea L.) is an important crop grown
worldwide. Its total yield in China has been the highest in
the world since 1993,1 with yield levels of 1.30 million tons
in 2007, 1.43 million tons in 2008, and 1.47 million tons in
2009.2 As previously reported, the protein content of peanuts
ranges from 154 to 302 g kg−1 , showing a large variation that
is greatly influenced by genotype and environment.3 Peanut is
the world’s third most important source of vegetable protein.4
Among vegetable proteins, only peanut protein does not
contain significant amounts of any intrinsic antinutritional factors,
making it a relatively inexpensive source of protein for human
consumption.5 In addition, research has shown that the nutritive
value of peanut protein is considerably better than that of wheat
protein.6
The protein and amino acid concentrations of peanuts are
important factors in peanut processing. Specifically, the total
protein concentration is crucial to peanut processing because
it determines how the peanuts can be processed into different
products such as roasted, candy and bars. Therefore, measurement
of peanut protein is a relevant component of processing. Knowing
the amino acid profile of protein as a quality parameter is also
important. The USA leads the world in direct consumption of
peanuts, and 70% of its peanut production is consumed as
food.7 Roasted peanuts are the most desirable peanut product
because of their pleasant and unique flavor, for which amino
acids and carbohydrates have been reported as precursors.8
118
sugars to contribute to the flavor quality of roasted peanut
seeds.9 Conventional methods used for total protein and amino
acid content determination could provide valuable information
regarding the variability of protein and amino acid concentrations
within a sample, but these methods are destructive and time
consuming when performed on large samples. In contrast,
nondestructive measurement of the kernel protein and amino
acids would result in considerable savings in both time and money
during grading and processing.
Near-infrared reflectance spectroscopy (NIRS) for food quality
measurements is currently being applied in food processing and
quality inspection, and it has several advantages over conventional
physical and chemical methods of analyzing food quality. Spectra
measured using NIRS contain absorption bands that are mainly
caused by three chemical bonds: C–H, which is usually from fats
and oil; O–H, which is found in water; and N–H, which is found
in protein.10,11 Other chemical bonds may appear in overtone
bands in the NIR region, but they are generally too weak to be
∗
Correspondence to: Qiang Wang, Institute of Agro-products Processing Science
and Technology, Chinese Academy of Agricultural Sciences; Key Laboratory of
Agro-product Processing and Quality Control, Beijing 100193, People’s Republic
of China. E-mail: wangqiang365@263.net
Institute of Agro-products Processing Science and Technology, Chinese
Academy of Agricultural Sciences; Key Laboratory of Agro-product Processing

Some researchers found that amino acids, including aspartic and Quality Control, Beijing 100193, People’s Republic of China

J Sci Food Agric 2013; 93: 118–124 www.soci.org

c 2012 Society of Chemical Industry
Contents of protein and amino acids in peanuts using reflectance spectroscopy
considered for analysis in such complex food systems as peanuts,
which contain water, oil, fat, and protein, among others. NIRS is
ideal for quantitatively determining oil, protein and moisture by
deducing C–H, N–H and O–H bonds. In addition, high scatter
coefficients allow for excellent diffuse reflectance spectra of solids
to be obtained.
NIRS may be applied with minimal sample preparation, and
it has been used in amino acid analysis by several researchers
with varying degrees of success. Van Kempen and Simmin12
evaluated NIRS for estimating the digestible amino acid contents
of several feed ingredients of animal origin. Cross-validation of
their calibration models for the prediction of lysine and methionine
resulted in determination coefficients (R2 ) ranging from 0.80 to
0.95. Williams et al.13 reported satisfactory results (R2 = 0.66–0.96)
in correlating NIRS data on ground wheat and barley with their
amino acid concentrations. Wu et al.14 showed the applicability
of NIRS to amino acid analysis in milled rice powder. In their
study, most of the amino acid calibration models had high
determination coefficients (R2 = 0.85–0.98), except for those
of cysteine (R2 = 0.78), histidine (R2 = 0.65), and methionine
(R2 = 0.10). Pazdernik et al.15 demonstrated that the accuracy of
NIRS screening for amino and fatty acid concentrations in soybeans
may be improved by grinding. Fontaine et al.16,17 obtained R2
values of 0.84–0.98 for soybeans and soybean meal. Overall, the
predictive ability of amino acid calibration models was found to
be dependent on the type of grain, sample form (whole grain
or ground), and specific amino acid. In peanut-related research,
Rao et al.18 reported a high determination coefficient (R2 = 0.97)
for peanut oil calibration models; Sundaram et al.19 assessed the
applicability of NIRS for peanut oil and fatty acid analysis in
peanuts nondestructively. They found that most of the fatty
acid calibration models had high residual predictive deviation
(RPD) values. Single-kernel devices were used to determine the
moisture and oil concentrations in nuts and grains by crushing a
kernel of a nut or grain from the bulk sample and measuring its
conductivity.20,21 Despite these advances in research, very little
is known about the use of the NIRS for protein and amino acid
analysis in peanuts.
Therefore, the aim of the present study was to evaluate the
ability of NIRS to predict the protein and amino acids contents of
peanut samples. Peanuts are a main source of plant protein for
humans. Modern diet formulation methods balance rations based
on amino acid contents, thereby increasing the need to develop
rapid and cost-effective techniques for protein and amino acid
measurement.
MATERIALS AND METHODS
Samples
One hundred and forty-one peanut samples were collected in 2011
from the following 12 provinces in China: Shangdong, 89 varieties;
Henan, 21 varieties; Guangdong, 8 varieties; Fujian, 5 varieties;
Guangxi, 4 varieties; Hubei, 4 varieties; Jiangsu, 5 varieties; Jiangxi,
1 variety; Liaoning, 1 variety; Hunan, 1 variety; Sichuan, 1 variety
and Hebei, 1 variety. Ninety-nine varieties were assigned to the
calibration set, whereas the remaining 42 varieties constituted the
validation set. All samples were stored at 4 ◦ C from collection until
analysis.
Chemical analysis
Samples were analyzed in duplicate and then averaged. Dry
matter (moisture) was determined by accurately weighing 2 g of
J Sci Food Agric 2013; 93: 118–124

c 2012 Society of Chemical Industry
www.soci.org
the sample in a tared glass, drying it in a ventilated oven for 4 h
at 103 ◦ C,22 cooling it afterwards in a desiccator and weighing
again. The crude protein (CP) content of the peanut samples was
measured using methods established by the American Association
for Cereal Chemists,23 and evaluated using a FOSS 2300 Nitrogen
Analyzer (FOSS, Sweden) and with a conversion factor of 5.46.
Aspartic (Asp), threonine (Thr), serine (Ser), glutamic (Glu), glycine
(Gly), leucine (Leu), arginine (Arg), and cysteine (Cys) were analyzed
and measured in our department using official method of the
American Association for Cereal Chemists.24
NIRS analysis
Peanut samples were scanned using a diode array analyzer (DA
7200, Perten Instruments, Huddinge, Sweden). Each sample (60 g)
was fitted in a 75 mm diameter cup that rotated during NIRS
scanning. Absorbance readings at 5 nm wavelength increments
were collected over an NIR wavelength range of 950–1650 nm.
Three scans were conducted on each sample and the data were
averaged before analysis.
Using multivariate regression software (Unscrambler, v. 9.7,
Camo, Oslo, Norway), partial least squares (PLS) (in which one
variable is modeled) was performed to develop prediction models
for protein and amino acid. Principal components analysis (PCA)
was performed before the application of PLS regression to reduce
the spectral data and derive the first nine principal components in
order to examine the possible grouping of samples and to detect
the spectral outliers as well.
Protein and amino acids were the dependent variables. PLS
regression analysis was conducted using the spectra from the
calibration dataset to develop an empirical equation for predicting
the concentrations of total protein and amino acids (Asp, Thr, Ser,
Glu, Gly, Leu, Arg, and Cys). The validation step was carried out
using the full cross-validation method and the external validation
method to ensure predictive ability and to avoid over-fitting of the
data.25 With cross-validation, the same samples were used for both
the calibration step and the validation step and the outlier samples
were not omitted. A sample was removed from the calibration
dataset, and the model was calibrated on the remaining data
points. The value for the removed sample was predicted, and
the prediction residual was computed. This process was repeated
with another sample in the calibration set until every object had
been removed once, after which all prediction residuals were
combined to compute the validation residual variance and root
mean square error of cross validation (RMSEV). The standard error
of calibration (SEC), the coefficient of determination in calibration
(R2c ), and RMSEV were calculated to evaluate the predictive ability
of the models. Moreover, the standard error of prediction (SEP),
bias, slope and the coefficient of determination in validation (R2v )
were calculated based on external validation to evaluate the
performance of models developed between the parameters of
predictive ability of the models developed for the whole sample
set. The ratio of the standard deviation of the validation set to the
standard error of validation or prediction (RPD) values was also
used to evaluate the goodness of fit. The RPD values usually range
from 1 to 10. Higher values indicate a stronger calibration model
for the accurate prediction of unknown sample composition. RPD
values of 1 or less are an indication of an inadequate model.
RPD values in the range 3.1–4.9 are good for initial screening
purposed (where an accurate prediction is not needed when a
large number of peanut batches are handled), and greater than
5 (range 5–6.4) is good for quality control and prediction.26 – 28
119
To develop NIRS calibration models for prediction of protein and
wileyonlinelibrary.com/jsfa
 www.soci.org L Wang et al.

Table 1. Statistics of the protein and amino acids in peanuts (n = 141) in the calibration and validation set and the repeatability (Sr) of reference
methods

Whole set Calibration set Prediction set

N Range Mean ± SD N Range Mean ± SD N Range Mean ± SD Sr

Crude protein 141 226.2–342.0 268.1 ± 19.6 99 237.1–342.0 282.5 ± 18.6 42 226.2–330.1 280.2 ± 22.5 5.40
ASP 141 22.6–49.0 33.4 ± 5.3 99 29.2–49.0 35.7 ± 3.7 42 22.6–45.6 29.8 ± 4.6 1.70
THR 141 5.4–9.6 7.3 ± 0.9 99 5.4–8.4 7.2 ± 0.5 42 5.4–9.6 7.4 ± 1.1 0.09
SER 141 8.8–20.8 13.8 ± 1.8 99 11.6–20.8 14.9 ± 1.6 42 8.8–18.3 12.0 ± 2.0 0.12
GLU 141 32.1–93.6 56.1 ± 10.6 99 46.4–93.6 63.1 ± 8.7 42 32.1–87.5 50.0 ± 11.9 2.34
GLY 141 10.3–23.6 16.1 ± 2.1 99 14.2–23.6 17.7 ± 1.5 42 10.3–21.6 14.5 ± 2.3 0.27
LEU 141 12.7–25.3 18.7 ± 1.5 99 15.7–25.3 18.9 ± 1.7 42 12.7–20.3 17.0 ± 2.1 0.85
ARG 141 22.8–47.3 33.5 ± 3.9 99 28.3–47.3 35.4 ± 3.6 42 22.8–46.2 30.6 ± 4.5 1.09
CYS 141 0.6–5.5 3.9 ± 0.7 99 2.5–5.5 4.0 ± 0.4 42 0.6–5.1 3.7 ± 1.0 0.13

amino acids, the entire (whole) dataset (141 samples) was split up
into a calibration set and a validation set. This experimental design
was chosen because a sample population could contain spectrally
similar samples whose characterization, in terms of the reference
analysis, could be expensive and unjustified.29
RESULTS AND DISCUSSION
Materials
Basic statistics of the protein and amino acid compositions for
the calibration and validation sets are summarized in Table 1.
Ninety-nine spectra were assigned to the calibration set, while
the remaining 42 constituted the validation set. The results
reveal a broad range in protein and amino acid contents,
especially in the calibration set, which is important for the
development of calibration models. However, the range in
the validation set was smaller for protein and most amino
acids. The precision of the chemical methods was assessed in
terms of the standard deviation of the difference between the
replicates (i.e. repeatability; Table 1). Repeatability was calculated
to evaluate the quality of the NIR predictive models. In addition,
the mean and standard deviation values indicated that the
formed sets were characterized by even constituent distributions,
suggesting that calibration sets will weight the calibration model
equally across the entire concentration range, with minimal
120
residuals at the extremes and relatively equal weighting at the
center.29
Chemical composition and NIRS analysis
Figure 1 shows the raw NIRS spectra of peanut samples over the
spectral range of 950–1650 nm. It contains information on the
relative proportions of C–H bonds (usually from fats and oil), O–H
bonds (found in water) and N–H bonds (found in protein).30 Five
absorption regions were observed (Fig. 2); the peaks at 950 nm
were probably related to second overtone O–H and N–H stretches
and third overtone C–H stretches; the peaks near 1120, 1180 and
1230 nm corresponded to the second overtone of C–H stretching
vibration, the combination band of O–H stretching and O–H
deformation, respectively; and the peaks around 1410 nm were
related to the first overtone of O–H and N–H in amino and amide
groups, which was the combination of the N–H stretching and
vibration with other vibration modes of the specific molecule.11,31
The second derived mathematical treatment was used to show the
five peaks probably related the second overtone O–H and N–H
stretches and the third overtone C–H stretches (Fig. 2). Figure 1
shows that the NIRS on peanuts with different varieties were
very similar to one another, and that the overlapping bands due
to overtones and combination modes makes the quantitative or
qualitative analysis not straight. Therefore, chemometric methods

Figure 1. NIRS of peanut samples (wave number, (nm) in abscissa and absorbance in ordinate).

wileyonlinelibrary.com/jsfa

c 2012 Society of Chemical Industry J Sci Food Agric 2013; 93: 118–124
Contents of protein and amino acids in peanuts using reflectance spectroscopy www.soci.org

E
B C D
A

Second-order derivative

Wavelength
Figure 2. The second-order derivative spectra of the peanut samples: (A) second overtone O–H and N–H stretches and third overtone C–H stretches; (B,
C, D) second overtone of C–H stretching vibration, the combination band of O–H stretching and O–H deformation, respectively; (E) first overtone of
O–H and N–H in amino and amide groups, which was the combination of the N–H stretching and vibration with other vibration modes of the specific
molecule.

were applied in this study to extract information from the spectra

for the analysis of protein and amino acids in the peanut samples.
The NIRS data represent a set of multiple variables that
contain overlapping information. Multivariate data analysis makes
it possible to extract useful information from original spectral
data, eliminate much overlapping information, and reduce the
dimension of data. PCA was used in this study. Details on the
PCA algorithm have been reported elsewhere.32 The principal
components (PCs) obtained from PCA were considered as
new eigenvectors of the original spectra. Samples with global
Mahalanobis distance (GH) < 3.0 from the mean of all spectra
were eliminated as spectral outliers (n = 3). In addition, samples
with neighborhood Mahalanobis distance (NH) smaller than 0.6
between the neighboring samples were eliminated from the initial
Figure 3. Score plot of the first two principal components (PC1 and PC2)
sample set in order to form the calibration sample set, whereas of peanut samples.
eliminated samples were used to create the external validation
sample set.
PCA was performed on the raw spectra across the full spectral used as reference method, the application of 2,4,4,1 mathematical
region (range 950–1650 nm) to examine the spectral variability treatment during calibration development using the calibration
of the selected samples. The score plot of the first two PCs of sample set resulted in optimal calibration and external validation
the raw spectra for peanut samples is shown in Fig. 3. The first statistics higher than those obtained after applying 4,10,10,1
two PCs explained more than 95% of the variations in the raw mathematical treatment to the whole sample set. These findings
spectra of peanut samples (PC1 91.05%, PC2 7.91%). Although the confirmed that the second derivative offers the most suitable
initial population was composed of peanut samples grown in 12 treatment for peanuts and that the fourth derivative could also be
localities, significant differences between the samples as well as very useful in spectral interpretation.34 Crude protein and amino
between peanuts were not observed (Fig. 3). acids were quantified using the PLS algorithm. For the calibration,
lower SEC and higher R2 values were considered better and more
Prediction of chemical composition accurate.35 Table 2 shows that the NIRS model developed for the
calibration sample set was significantly higher in relation to the
Peanut samples were divided into two subsets: the calibration
predictive ability of models developed for the whole sample set
set was used to build models, whereas the validation set was
(P < 0.001).
used to test their robustness. Prior to PLS regression, standard
normal variate and detrending (SNV-DT) scatter correction was
applied. Twenty-two mathematical treatments (D,G,S,S), varying Crude protein calibration and prediction
in terms of the order of the derivative (D), the gap over which the Calibration, cross-validation, and external validation statistics
derivative was calculated (G), and the number of data points used for NIRS model for different sample sets are shown in Table 2,
to smooth the data (S,S), were tested during the development which summarizes the performance parameters obtained for the
of the NIRS calibrations, which resulted in 22 calibration models. calibration equations. The calibration model developed using
As no mathematical treatment to which the spectral data were reflectance gave an R2 value of 0.99 for peanuts with low RMSEV
subjected in the process of calibration development could be values (0.19); RPD values were calculated to find the goodness
suggested as better than any other, the iterative procedure of trial of fit of the calibration model, with a value of 6.53 obtained
and error was performed; this resulted in 22 calibration models with for crude protein showing that this model should work well for
121

different calibration features.33 When the chemical method was quality control and analysis. Calibration and validation equations

J Sci Food Agric 2013; 93: 118–124

c 2012 Society of Chemical Industry wileyonlinelibrary.com/jsfa
 www.soci.org L Wang et al.

Table 2. Calibration and validation statistics of the NIRS model for protein and amino acid content prediction developed for the sample set

Calibration Validation

Parameter Math treatment R2c RMSE R2 RMSEV SEP R2v Bias Slope RPD

Crude protein 2,4,4,1 0.99 0.19 0.98 0.29 0.03 0.92 0.00 1.000 6.53
ASP 0.88 0.18 0.6 0.34 0.21 0.84 −0.033 1.106 2.52
THR 0.83 0.03 0.77 0.04 0.03 0.82 0.006 1.023 3.00
SER 0.86 0.09 0.54 0.17 0.1 0.82 −0.011 1.11 2.40
GLU 0.87 0.45 0.54 0.86 0.49 0.85 0.063 1.085 2.57
GLY 0.88 0.09 0.55 0.18 0.11 0.82 −0.003 1.047 2.36
LEU 0.88 0.09 0.61 0.17 0.05 0.81 −0.033 0.983 3.00
ARG 0.89 0.16 0.66 0.29 0.17 0.86 −0.028 1.059 2.88
CYS 0.96 0.01 0.94 0.02 0.004 0.995 0.000 0.992 7.50

for amino acids were suitable for quality assurance applications.

34 Protein NIRS calibration models with R2 ranging from 0.83 to 0.90 were
categorized as models ‘usable with caution for most applications’.
32
Y=0.27637+0.99006x The results implied that measuring amino acids in peanuts by
NIRS combined with multivariate data analysis is feasible. Figure 5
NIRS Predicted

30
28
26
24
22
22 24 26 28 30 32
measured value
Figure 4. References measured versus NIR predicted by the PLS model in
the calibration set and prediction set.
for protein content showed that NIRS and reference values were
closely associated. A calibration plot between predicted values
and measured values for total protein is shown in Fig. 4, which
illustrates excellent accuracy. The goodness of the calibration
for protein content was comparable with calibration equations
previously developed for seed protein in rapeseed36 and in single
seeds of wheat and soybean.37 The coefficient of determination
in the cross-validation of protein calibration developed in the
present study was excellent (R2 = 0.99) and compares well with
the results of Orman and Schumann,38 who used three types
of spectral data to predict the content of protein in maize
grain; their study showed that the protein values predicted
by equations developed from all three types of spectral data
correlated well with reference values, having R2 values ranging
from 0.83 to 0.98.
Amino acid calibration and prediction
Calibration, cross-validation, and external validation statistics for
NIRS models for different sample sets are shown in Table 2, which
shows the comparative results of amino acid analysis from the
PLS model based on PCA feature extraction. Overall, R2 values for
different amino acids ranged from 0.83 to 0.96 and RPD ranged
from 2.35 to 7.50. Based on the guidelines for interpreting R2
122
shows the scatter plot of references measured and predicted
using NIRS with PLS models in the prediction set. The prediction of
the amino acid content of peanuts using PLS regression analysis
developed as a standard showed a high coefficient between
the values of the reference method and the NIRS prediction.
Acceptable correlations between the predicted values and the
reference values were also found for Met, although the coefficients
of determination were of a significantly lower R2 (0.49) (data not
shown).
In addition, linear regression between amino acid and crude
34 protein (CP) contents for the same sample populations was
calculated (Table 3). As a consequence, the r of the linear
regression of amino acids to CP was mostly high for the sample
population and equal to or slightly lower than the results from
NIRS calibration. Indeed, Thr NIRS explained much more of the
variance (0.91) than the CP regression, with a poor correlation of
0.83. Rubenthaler and Bruinsma40 were the first to report amino
acid prediction using NIRS. They calibrated the ratio of Lys to CP
for several small wheat populations and obtained coefficients of
correlation, r, between 0.85 and 0.98. They concluded that NIRS
predicted amino acids independent of CP. Workman41 presented
calibration data based on 111 calibration samples that were
selected to avoid spectral similarities from 400 corn samples.
Using equipment and a calibration algorithm similar to those used
in our laboratory, Workman41 achieved R2 values between 0.62
and 0.89 for 12 amino acids. The RMSEVs obtained were in the
range 0.02–0.14. Validation with 30 independent samples with
R2 values of 0.23–0.58 gave very low correlations for five amino
acids and an SEP of 0.01 (Cys and Trp) or 0.02 (Met, Lys, and
Thr). With data for soybean meal as well, Workman41 nevertheless
concluded that NIRS is a highly promising method for rapid amino
acid measurement in major feed ingredients. Dyer and Feng42,43
used NIR amino acid calibrations for screening purposes in the
development of genetically altered grains. Based on data from
150 corn samples, they reported the following statistical data:
R2 = 0.78 and RMSEV = 0.012 for methionine; RMSEV = 0.013 for
Cys; R2 = 0.93 and RMSEV = 0.017 for Lys and RMSEV = 0.013 for
Thr. Despite good correlations, the standard errors were mostly
higher than those obtained in our laboratory based on very

outlined by Williams and Norris,39 the NIR calibration equations accurate reference analysis.

wileyonlinelibrary.com/jsfa

c 2012 Society of Chemical Industry J Sci Food Agric 2013; 93: 118–124
Contents of protein and amino acids in peanuts using reflectance spectroscopy www.soci.org

5.0 1.0
ASP Thr

4.5 0.9
Y=0.44984+0.86531x Y=0.12201+0.83301x

NIRS Predicted
NIRS Predicted
4.0
0.8
3.5
0.7
3.0
0.6
2.5

2.0 0.5
2.0 2.5 3.0 3.5 4.0 4.5 5.0 0.5 0.6 0.7 0.8 0.9 1.0
Measured value Measured value
10
Glu
Ser
2.0 9
Y=0.75807+0.86947x
1.8 Y=0.18891+0.86298x 8

NIRS Predicted
NIRS Predicted

1.6 7

1.4 6

1.2 5

1.0 4

0.8 3

0.8 1.0 1.2 1.4 1.6 1.8 2.0 3 4 5 6 7 8 9

Measured value Measured value

2.8 Leu
Gly
2.4
2.6
Y=0.24324+0.85234x
2.2 Y=0.13165+0.92507x
2.4
2.0

NIRS Predicted
NIRS Predicted

2.2
1.8 2.0
1.6 1.8
1.4 1.6
1.2 1.4
1.0 1.2

1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6
Measured value Measured value
5.0 Arg
0.6 Cys
4.5
Y=0.36451+0.89119x 0.5 Y=0.0149+0.96138
NIRS Predicted
NIRS Predicted

4.0
0.4
3.5
0.3

3.0 0.2

2.5 0.1

2.0 0.0
2.0 2.5 3.0 3.5 4.0 4.5 5.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6
Measured value Measured valued

Figure 5. References measured versus NIR predicted by the PLS Model in the calibration set and prediction set.

Table 3. Linear regression of amino acids to crude protein (CP)

CONCLUSIONS
This study evaluated protein and amino acids in peanuts from
Linear regression of amino acids China using NIRS, with the results showing that the protein and
to crude protein (CP)
amino acid contents of peanuts can be determined by NIRS
Parameter Intercept Slope r coupled with PLS calibration models. Its acceptable predictive
accuracy and the simplicity of measurement make this method
ASP 25.16 0.82 0.86 valuable to industrial (particularly online) applications, for which
THR 28.29 0.778 0.83 further studies are required. However, improvements in the
SER 25.87 1.45 0.87 prediction precision of calibration models by using a higher
GLU 26.62 0.2 0.85 number of samples and other chemometric methods are also
GLY 26.3 0.917 0.90 needed. In summary, the results of this study indicate that NIRS
LEU −3.39 16.98 0.89 calibration equations developed as a standard are reliable in
ARG 22.17 1.761 0.93 predicting the contents of total protein and amino acids in
CYS 22.26 14.76 0.95 peanuts.

These results indicate that determining crude protein and amino ACKNOWLEDGEMENTS
acids using NIRS in peanuts is possible. Near-infrared regions This work was supported by the ‘Special Fund for Agro-scientific
have considerable influence on the spectra owing to the strong Research in the Public Interest’ (Serial No. 200903043). We thank
relationship between protein and amino acids, mainly with O–H our academic colleagues for many stimulating discussions in this
123

overtones, C–H combination tones, and N–H overtones.44 field.

J Sci Food Agric 2013; 93: 118–124

c 2012 Society of Chemical Industry wileyonlinelibrary.com/jsfa
 www.soci.org
REFERENCES
1 FAOSTAT (2009). [Online]. Available: http://faostat.fao.org/site/567/
DesktopDefault.aspx?PageID=567#ancor [9 May 2012].
2 China Statistical Yearbook (2010). [Online]. Available: http://www.stats.
gov.cn/tjsj/ndsj/2009/indexch.htm [9 May 2012].
3 Afolabi OA, Ojofeitimi EO and Oke OL, Chemical and clinical evaluation
of groundnut maize gruel mixture (Epa-ogr) in the amelioration of
protein energy malnutrition in the developing countries. Nutr Rep
Int 38:621–628 (1989).
4 Lusas EW, Food uses of peanut protein. J Am Oil Chem Soc 56:425–530
(1979).
5 Prakash V and Rao MS, Physicochemical properties of oilseed proteins.
CRC CR Rev Biochem Mol 20:286–301 (1986).
6 Bodwell CE, The nutritive value of the same proteins preparations as
estimated by human, rat and chemical assays. J Am Oil Chem Soc
56:156–160 (1979).
7 Agricultural Statistics. US Government Printing Office, Washington, DC
(1977).
8 Pickett TA and Holley KT, Peanut Roasting Studies. Georgia Agricultural
Experimental Station Technical Bulletin No. 1 (1952).
9 Cobb WY and Swaisgood HE, Roasted peanut flavor and its relation to
growth environment. J Food Sci 36:538–539 (1971).
10 Cozzolino D, Kwiatkowski MJ, Dambergs RG, Cynkar WU, Janik LJ
and Skourou G, Analysis of elements in wine using near
infrared spectroscopy and partial least squares regression. Talanta
74:711–716 (2008).
11 Osborne BG and Fearn T, Near Infrared Spectroscopy in Food Analysis.
Longman Scientific and Technical, Harlow, UK, pp. 20–42 (1986).
12 Van Kempen TATG and Simmins PH, Near-infrared reflectance
spectroscopy in precision feed formulation. J App Poult Res
6:471–477 (1997).
13 Williams PC, Preston KR, Norris KH and Starkey PM, Determination
of amino acids in wheat and barley by near-infrared reflectance
spectroscopy. J Food Sci 49:17–20 (1984).
14 Wu JG, Shi C and Zhang X, Estimating the amino acid composition in
milled rice by near-infrared reflectance spectroscopy. Field Crop Res
75:1–7 (2002).
15 Pazdernik DL, Killam AS and Orf JH, Analysis of amino and fatty
acid composition in soybean seed using near infrared reflectance
spectroscopy. Agron J 89:679–685 (1997).
16 Fontaine J, Horr J and Schirmer B, Near-infrared reflectance
spectroscopy enables the fast and accurate prediction of the
essential amino acid contents in soy, rapeseed meal, sunflower
meal, peas, fishmeal, meat meal products, and poultry meal. J Agric
Food Chem 49:57–66 (2001).
17 Fontaine J, Schirmer B and Horr J, Near-infrared reflectance
spectroscopy (NIRS) enables the fast and accurate prediction of
the essential amino acid contents. 2. Results for wheat, barley, corn,
triticale, wheat bran/middlings, rice bran, and sorghum. J Agric Food
Chem 50:3902–3911 (2002).
18 Rao Y, Xiang B, Zhou X, Wang Z, Xie S and Xu J, Quantitative and
qualitative determination of acid value of peanut oil using near-
infrared spectrometry. J Food Eng 93:249–252 (2009).
19 Sundaram J, Kandala CV, Holser RA and Butts CL, Determination of
in-shell peanut oil and fatty acid composition using near-infrared
reflectance spectroscopy. J Am Oil Chem Soc 87:1103–1114 (2010).
20 Kandala CVK and Nelson SO, Measurement of moisture content in
single kernels of peanuts: a nondestructive electrical method. Trans
ASAE 33:567–572 (1990).
21 Kandala CVK, Moisture determination in single peanut pods by
complex RF impedance measurement. IEEE Trans Instrum Meas
53:1493–1496 (2004).
22 VDLUFA, Methodenbuch Band III. Die chemische Untersuchung von
Futtermitteln, 4th suppl. VDLUFA Verlag, Darmstadt (1997).
23 AOAC International, Method 950.48. Rutin in drugs, in Official Methods
of Analysis. AOAC, Arlington, VA (2011).
124
wileyonlinelibrary.com/jsfa

c 2012 Society of Chemical Industry
L Wang et al.
24 AOAC International, Method 994.12. Rutin in drugs, in Official Methods
of Analysis. AOAC, Arlington, VA (2011).
25 Martens H and Næs T, Multivariate Calibration. Wiley, New York, p. 419
(1989).
26 Williams PC, Implementation of near-infrared technology, in Near-
Infrared Technology in the Agricultural and Food Industries, ed. by
Williams PC and Norris KH. American Association of Cereal Chemists,
St Paul, MN, pp. 145–171 (2001).
27 Fearn T, Assessing calibration; SEP, RPD, RER and R2 . NIRNews 13:12–14
(2000).
28 Malley DF, McClure C, Martin PD, Buckley K and McCaughey WP,
Compositional analysis of cattle manure during composting using
a field-portable near-infrared spectrometer. Commun Soil Sci Plant
36:455–475 (2005).
29 Shenk JS and Westerhaus MO, Population structuring of near infrared
spectra and modified partial least squares regression. Crop Sci
31:1548–1555 (1991).
30 Hourant P, Baeten V, Morales MT, Meurens M and Aparicio R, Oil and
fat classification by selected bands of near infrared spectroscopy.
Appl Spectrosc 54:1168–1174 (2000).
31 Shenk J, Westerhous M and Workman J, Application of NIR
spectroscopy to agricultural products, in Handbook of Near Infrared
Analysis, ed. by Ciurczak E. Practical Spectroscopy Series. Marcel
Dekker, New York, pp. 383–431 (1992).
32 Serudo RL, Oliveira LC, Rocha JC, Paterlini WC, Rosa AH, Silva HC, et al,
Reduction capability of soil humic substances from the Rio Negro
basin, Brazil, towards Hg(II) studied by a multimethod approach
and principal component analysis (PCA). Geoderma 138:229–236
(2007).
33 Barton FE II, New methods for the structural and compositional analysis
of cell walls for quality determinations. Anim Feed Sci Technol
32:1–11 (1991).
34 ISIWindows Near Infrared Software, WinISI III Manual, version 1.60.
Infrasoft International, State College, PA (2005).
35 Windham WR, Roberson JA and Leffler RG, A comparison of methods
for moisture determination of forages for near infrared reflectance
spectroscopy calibration and validation. Crop Sci. 27:777–783
(1986).
36 Velasco L and Becker HC, Estimating the fatty acid composition of
the oil in intact seed rapeseed (Brassica napus L.) by near infrared
reflectance spectroscopy. Euphytica 101:221–230 (1998).
37 Abe H, Kusama T, Kawano S and Iwamoto M, Nondestructive
determination of protein content in a single kernel of wheat
and soybean by near infrared spectroscopy, in Near Infrared
Spectroscopy: The Future Waves, ed. by Davies AMC and Williams.
NIR Publications, Chichester, pp. 457–461 (1995).
38 Orman BA and Schumann RA, Comparison of near-infrared
spectroscopy calibration methods for the prediction of protein,
oil, and starch in maize grain. J Agric Food Chem 39:883–888 (1991).
39 Williams P and Norris K (eds), Near-Infrared Technology in the
Agricultural and Food Industries (2nd edn). AACC, St Paul, MN (2001).
40 Rubenthaler GL and Bruinsma BL, Lysine estimation in cereals by NIR.
Crop Sci 18:1039–1042 (1978).
41 Workman J, NIRS measurement of amino acids in feed ingredients, in
Proceedings of the Georgia Nutrition Conference for the Feed Industry,
University of Georgia, Athens, GA, pp. 37–47 (1991).
42 Dyer DJ and Feng P, NIR applications in the development of genetically
altered grains, in Near Infrared Spectroscopy: Proceedings of the 7th
International Conference. NIR Publications, Chichester, pp. 490–493
(1996).
43 Dyer DJ and Feng P, NIR destined to be major analytical influence.
Feedstuffs 10:24–25 (1997).
44 Garnsworthy PC, Wiseman J and Fegeros K, Prediction of chemical,
nutritive and agronomic characteristics of wheat by near infrared
spectroscopy. J Agric Sci 135:409–417 (2000).
J Sci Food Agric 2013; 93: 118–124