Food Chemistry 127 (2011) 275281

Contents lists available at ScienceDirect

Food Chemistry
journal homepage: www.elsevier.com/locate/foodchem

Analytical Methods

Prediction of the amino acid composition in brown rice using different

sample status by near-infrared reectance spectroscopy
B. Zhang a,1, Z.Q. Rong a,1, Y. Shi a, J.G. Wu b,, C.H. Shi a,
a
b

Department of Agronomy, College of Agriculture and Biotechnology, Zhejiang University, Hangzhou 310029, PR China
Institute of Crop Science, College of Agriculture and Biotechnology, Zhejiang University, Hangzhou 310029, PR China

a r t i c l e

i n f o

Article history:
Received 29 May 2009
Received in revised form 20 December 2010
Accepted 29 December 2010
Available online 8 January 2011
Keywords:
Brown rice
Calibration equation
Near-infrared reectance spectroscopy
(NIRS)
Amino acid
Foodstuff

a b s t r a c t
In this study, 279 samples of brown rice grains and their our, selected from a larger original population,
were scanned by NIRSystem model 5000 mono-chromator in these two kinds of sample status for nearinfrared reectance spectroscopy (NIRS) analysis. Spectral pretreatment method 2,8,8,1 combined with
SNV + D scatter correction was found suitable for developing calibration equations for amino acids. Equations for total amino acid content and for all individual amino acids, excluding cystine, methionine and
tyrosine, were developed with this spectral pretreatment method. These equations had low SECV (0.010
0.063%) and SEP (0.0110.066%); with high 1 VR (0.8780.960), R2 (0.8370.947) and SD/SEP (2.421
4.333). The results suggest that equations for the thirteen amino acids from the two sample categories
can be directly used to estimate the amino acid composition in brown rice. This indicates once more that
NIRS is a powerful technology that could be very useful for the determination of amino acid content in
breeding programs that involve brown rice as well as for quality control in the food industry.
2011 Elsevier Ltd. All rights reserved.

1. Introduction
Rice (Oryza sativa L.) is the most important cereal crop worldwide and especially in Asian countries due to its good taste and
nutritional value. Rice quality in terms of nutrition is valuable for
its protein content and the balance of essential amino acids (Mitchell & Block, 1946). Some studies have revealed that the outer portion of the grain contains various compounds including many
important nutritional and bio-functional factors (Itani, Tamaki,
Arai, & Horino, 2002; Lamberts et al., 2007; Liang et al., 2008). Rice
protein is also an excellent substitute for powdered milk because
of its low hyper susceptibility for babies. With the progress in food
processing technology, different novel products such as breakfastcereals, brown rice syrup, brown rice tea (Ohtsubo, Suzuki, Yasui, &
Kasumi, 2005) and brown rice noodles have been made. Food products made from brown rice would be more acceptable to consumers and the food industry since they have more nutritional and biofunctional components than ordinary rice products (Ohtsubo et al.,
2005). Brown rice also has a great potential in production of whole
grain food products because of its high nutritional and health benets. A rapid analytical method is required for selection of breeding
materials as well as for quality control during brown rice process Corresponding authors. Tel./fax: +86 571 86971691 (J.G. Wu), +86 571
86971117 (C.H. Shu).
E-mail addresses: jgwu@zju.edu.cn (J.G. Wu), chhshi@zju.edu.cn (C.H. Shi).
1
These authors contributed equally to this paper.
0308-8146/$ - see front matter 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.foodchem.2010.12.110

ing. This technology could help develop new cultivars of brown

rice with superior quality and also produce new kinds of processed
foods having high quality protein and amino acids.
Near-infrared reectance spectroscopy (NIRS) has the advantage of being safe, efcient, more economical and environment
friendly. It can be used for the simultaneous determination of several traits with a single measurement (Shenk & Westerhaus, 1993).
It is also ideal for rapid screening in quality detection systems and
plant breeding programs (Wu & Shi, 2003). NIRS had been widely
applied to the prediction of quality traits in milled rice grains such
as amylose content (Bao, Cai, & Corke, 2001; Shu, Wu, Xia, Gao, &
McClung, 1999; Wu & Shi, 2004, 2007), gelatinization temperature,
gel consistency and alkali spread value (Bao et al., 2001; Delwiche,
McKenzie, & Webb, 1996; Wu & Shi, 2007), fatty acids (Li & Shaw,
1997), protein content (Delwiche et al., 1996; Li et al., 2006; Xie,
Xiao, Li, & Li, 2006) and retrogradation properties (Bao, Shen, &
Jin, 2007). Furthermore, NIRS enables fast and accurate prediction
of the essential amino acid contents in soy, rapeseed meal, sunower meal, peas, shmeal, meat meal products, and poultry meal
(Fontaine, Hrr, & Schirmer, 2001). Pazdernik, Killam, and Orf
(1997) tried to establish the NIRS technique/model for analyzing
amino acid contents in soybean and found it useful as a gross
screening method for improving many amino acids contents in
soybean seed. Gonzlez-Martn, lvarez-Garca, and GonzlezCabrera (2006) developed the equations for determining the amino
acid contents in animal feed, and indicated that the amino acid
values for feed predicted with NIRS using a ber optic probe are

276

B. Zhang et al. / Food Chemistry 127 (2011) 275281

comparable to those determined with the chemical ion-exchange

HPLC method. Wu, Shi, and Zhang (2002) developed NIRS calibration equations to predict the amino acid and nitrogen content of
milled rice powder, and found that equations for 15 different amino acids, except for cysteine, methionine and histidine, showed
high coefcients of determination and low standard errors in calibration. Hitherto, no research effort has been made to determine
amino acid composition in brown rice using NIRS. The purpose of
the present study was to develop the NIRS model for the determination of amino acid composition in brown rice and optimize the
calibration equations for rice breeding programs and for quality
control in the food industry.
2. Materials and methods
2.1. Materials
A total number of 518 indica rice samples were collected as original sample population, and analyzed with the NIRS method to
develop and validate various calibration equations. All the samples
were breeding materials with high variation for agronomic traits;
selected from a wide range of rice germplasm including varieties
and advanced lines that had grown under different environmental
conditions for three years from 2004 to 2006.
Rice grain samples were dehusked with an electrical dehusker
(model B-76, Huangyan, Zhejiang Province, China). Parts of brown
rice were further ground by cyclone grinder (equipped with
0.5 mm screen) with model number 3010-019 (Fort Collins, Colorado, USA) and prepared for NIRS scanning and chemical analysis.
2.2. Spectrum collection and sample selection
Two replications, each of the brown rice and brown rice our
samples were scanned on a NIRS model 5000 monochromator
(NIRSystem, Silver Spring, MD, USA) equipped with sample autochanger. Both brown rice and brown rice our samples were
placed in a small ring cup of 36 or 18 mm inner diameter, respectively, and reectance spectra (log 1/R) from 1100 to 2498 nm
were recorded at 2 nm intervals. Each spectrum represented the
average of 32 scans, and was recorded as log (1/R) (Shenk & Westerhaus, 1993). Each sample was scanned twice by rotating the ring
cup to a different position in order to minimize the effects of particle size. The average spectrum of each sample was calculated for
further chemometric analysis. The spectroscopic procedures and
data recording were conducted in the WinISI II software (v1.5,
FOSS NIRS, Silver Springs, MD, USA).
To establish a good regression equation, population denition
and sample selection were necessary. For population denition,
the CENTER algorithm (Shenk & Westerhaus, 1991a) was applied
to calculate Mahalanobis distances (so-called Global H distance,
GH). The scatter of spectra were rst corrected by a standard normal
variate and de-trending (Barnes, Dhanoa, & Lister, 1989) and derivative with a 1,4,4,1 math treatment. In the procedure of principal
components analysis (PCA), the loading type of PCA was used when
the variations were limited to the spectra only. In the present study,
the population was dened through a cutoff of GH = 3 by using CENTER program, resulting in the new population of 514 samples after 4
abnormal or extreme samples which appeared as outliers were removed. In order to select the suitable and representative samples
for the calibration set, the SELECT algorithm (Shenk & Westerhaus,
1991a) was applied, and the pretreatment method for spectra were
done as in the CENTER program. Then Mahalanobis distances between all pairs of spectra (neighbor H distance, NH) were calculated.
Calibration set was constructed by a cutoff of NH = 0.35, resulting in
the 279 selected samples dispersed in the global space, which were
used to develop the calibration equation.

2.3. Determination of amino acid

The our samples were further analyzed for reference with HCl
hydrolyzed HPLC methodology. In order to balance the moisture
content, all samples were placed in an airtight box for three days.
After that, the moistures of the samples were detected at a level of
10.18%. About 50 mg of each sample was weighed with a readability of 0.1 mg and added into a test tube with 5 ml of 6 M HCl. The
test tube was vacuumized with a vacuum pump, airproofed with
an alcohol blowtorch, and then placed into a constant temperature
box. The hydrolysis was maintained for 24 h at 110 C. The test
tube was removed and opened after cooling; and the diluted sample was transferred to a 50 ml capacity bottle and adjusted to
50 ml. Ten ml of the dilution was transferred to a distillation ask
with water bath at 60 C, which was held to rotary evaporator
(model RE-52AA, Shanghai, China) for evaporation. When the solution was evaporated, the residue was dissolved in 10 ml 0.02 M HCl
and ltered with 0.22 lm lm. About 1 ml of each sample liquid
was injected into an auto-sampler bottle and placed in an amino
acid auto-analyzer model L-8900 (Hitachi High-Technologies Corporation, Japan).
The amount of each amino acid in the samples was calculated
with reference to the standard sample by the EZChrom Elite (Hitachi High-Technologies Corporation, 2004) software, and each
amino acid content was expressed as a percentage of the total sample weight. Determinations were made for each sample for amino
acid with two replications.
2.4. Calibration and external validation
Calibration and validation were conducted for brown rice and
its our spectra respectively following the manual of the WinISI
II software. The population was divided into calibration set and
external validation set. All the samples were chosen by the pattern
of one of every 6 samples for the external validation set and the
remaining samples were used as the calibration set. The calibration
set was used to develop the calibration equation between spectral
data and laboratory reference values through different mathematical treatments, and external validation set was used to evaluate
the calibration equation. To facilitate comparison between the calibration models developed from grain and our spectra, only the
our samples were divided manually into two equal sets. Calibration equations were developed by using the WinISI II 1.5 software
(WINISI II, Infrasoft International, LLC, Port Matilda, PA, USA) in the
spectral information range of 11002498 nm. Four different math
treatments (D = 1, G = 4, S1 = 4, S2 = 1; 1,8,8,1; 2,4,4,1 and 2,8,8,1)
were applied for calibration (Barnes et al., 1989). D is the derivative
order number. It is a good way of solving problems associated with
overlapping peaks and baseline correction by using derivative
spectra instead of the raw optical data to perform calibration
(Hruschka, 1987). G is the gap (the number of data points over
which the derivation is computed), S1 is the number of data points
in the 1st smoothing and S2 is the number of data points in the 2nd
smoothing which is normally set at 1 if there is no 2nd smoothing.
In addition to derivatives, standard normal variate and de-trending
(SNV + D) algorithms (Barnes et al., 1989) were applied to the derived spectra to minimize baseline offset due to scattering effects
caused by differences in particle size or path-length variation
among samples.
Cross validation was used to prevent over-tting (Shenk &
Westerhaus, 1993). With the number of parameters set to default and the number of cross-validation groups set to approximately n/2 with a maximum of 30 (Shenk & Westerhaus, 1991b),
modied partial least squares (MPLS) were used to develop the
regression equations. Two cycles of outlier elimination were set
up with samples with an H value (Mahalanobis distance) larger

277

B. Zhang et al. / Food Chemistry 127 (2011) 275281

study were representing a high level of diversity in brown rice.

The broad variability could be due to the use of diverse rice genotypes, growth years and growth locations for samples, both in the
calibration set and external validation set, which was important
for developing calibration equations to be used for future predictions (Gonzlez-Martn et al., 2006; Lu, Huang, & Zhang, 2006).
Furthermore, the small differences in means, ranges and standard
deviations between the calibration set and the validation set indicated that calibration and external validation sets were selected
properly and both sets could be representative of the overall variations in rice samples, thus making the population suitable for
NIRS calibration.

than 4 (spectral outlier) and a T value larger than 2.5 (sample

which did not t the calibration model) being eliminated (Shenk
& Westerhaus, 1993). The accuracy of the calibration model was
characterized by various NIRS performance values, e.g. standard error of calibration (SEC) and the coefcient of determination (RSQ)
for calibration, coefcient of determination (1 VR) and standard
error of cross-validation (SECV) for cross-validation (Shenk &
Westerhaus, 1996). The best calibration equations for this type of
data were judged by the highest calibration RSQ (or statistic
1 VR as an estimate of the coefcient of determination), and
the lowest SEC or SECV (Wu et al., 2002).
The application of only cross-validation to assess the calibration
equations might not be sufcient. A subsequent external validation
of the initial calibration model using samples independent from
the calibration set led to further NIRS performance values for each
constituent. The standard error of performance (SEP) and the coefcients of determination (R2) could well describe the NIRS analytical error when analyzing samples of unknown quantitative
composition. The prediction ability of each equation was tested
based on the following statistics: coefcient of determination
(R2) and standard error of performance (SEP). The ratio of SD/SEP
(Shenk & Westerhaus, 1996) was also used to evaluate the precision of the NIRS equations.

3.2. Selection of spectra pretreatment methods

In order to ascertain which mathematical treatment with scatter correction was better to develop calibration equation of brown
rice our, three kinds of amino acids, i.e. glutamic acid, phenylalanine and arginine were randomly selected to develop calibration
equations with different mathematical treatments (0,0,1,1;
1,4,4,1; 1,8,8,1; 2,4,4,1; 2,8,8,1) and scatter correction (SNV + D).
Prediction equations for these amino acids were developed by
modied partial least squares regression and evaluated by the
external validation.
The results in Table 2 showed the effects of different mathematical treatments with SNV + D for calibration equations with brown
rice our samples (about 3 g) scanned by using full cup. For the
equations of glutamic acid, the pretreatment of 1,4,4,1 and
2,8,8,1 with each combined with SNV + D were much better than
the control (0,0,1,1), with the latter one being the best. The other
combination i.e. pretreatments or only scatter correction was
found to be less precise than control in external validation. For
the equations of phenylalanine, the pretreatment method of
2,8,8,1 and SNV + D had a better effect than the control, whereas
the others were similar to the control based on external validation
even though their calibration effects seemed to be better than that
of the control. For the equations of arginine, the pretreatment
method of 1,4,4,1, 2,4,4,1 and 2,8,8,1 with SNV + D had better effect
than the other pretreatment methods based on the external validation. In brief, all prediction equations with different mathematical
treatments were excellent, showing high coefcients of determination (RSQ: 0.9200.969; 1 VR: 0.8800.960; R2: 0.8680.946),
and low standard errors (SEC: 0.0130.067%, SECV: 0.0140.069%,

3. Results
3.1. Variability for amino acid composition
A total of 279 samples, including 230 samples for calibration
and 49 samples for external validation, were analyzed for 17 different amino acids with the HCl hydrolyzed HPLC method. Table 1
shows the statistics of amino acid composition in brown rice in
the calibration set and external validation set. There were three
levels of amino acid contents in brown rice i.e. high levels with
more than 0.866% aspartic acid, arginine, leucine, glutamic acid;
intermediate level between 0.493% and 0.693% in serine, alanine,
phenylalanine, valine, glycine, and low level of less than 0.434% in
threonine, proline, tyrosine, lysine, histidine, isoleucine, cysteine
and methionine. Large SD ranges from 0.036% to 0.286% were observed in amino acids. Total amino acid content (TAA) covered a
wide range between 6.928% and 15.769% with a high SD of
1.329% in the calibration set, which reects a broad range of
variability. These ndings suggest that the samples used in this

Table 1
Amino acid composition of brown rice for the NIRS calibration and external validation set.
Constituent

Aspartic acid
Threonine
Serine
Glutamic acid
Glycine
Alanine
Cysteine
Valine
Methionine
Isoleucine
Leucine
Tyrosine
Phenylalanine
Lysine
Histidine
Arginine
Proline
TAA
a

SD: standard deviation.

Calibration

External validation
a

Min (%)

Max (%)

Mean (%)

SD

0.673
0.265
0.339
1.358
0.340
0.427
0.084
0.439
0.127
0.282
0.572
0.246
0.366
0.282
0.151
0.599
0.295
6.928

1.609
0.619
0.760
2.982
0.807
1.016
0.236
0.974
0.292
0.65
1.279
0.63
0.827
0.763
0.426
1.399
0.701
15.769

0.983
0.391
0.520
2.096
0.493
0.632
0.151
0.649
0.209
0.434
0.866
0.403
0.573
0.426
0.266
0.911
0.434
10.434

0.127
0.047
0.067
0.286
0.058
0.081
0.024
0.080
0.027
0.057
0.115
0.067
0.077
0.055
0.036
0.122
0.054
1.329

(%)

Min (%)

Max (%)

Mean (%)

SD (%)

0.756
0.299
0.392
1.577
0.394
0.482
0.091
0.498
0.071
0.321
0.645
0.275
0.424
0.300
0.159
0.705
0.324
7.941

1.299
0.499
0.702
2.937
0.630
0.848
0.206
0.853
0.274
0.597
1.207
0.503
0.773
0.525
0.345
1.152
0.542
13.799

0.982
0.393
0.523
2.110
0.492
0.632
0.150
0.651
0.207
0.434
0.870
0.403
0.574
0.425
0.266
0.910
0.431
10.454

0.105
0.039
0.058
0.247
0.048
0.071
0.024
0.068
0.035
0.050
0.104
0.054
0.067
0.046
0.033
0.099
0.041
1.119

278

B. Zhang et al. / Food Chemistry 127 (2011) 275281

Table 2
Comparisons of the effects of different mathematical treatments with scatter corrections of standard normal variance and de-trend (SNV + D) for calibration equations of brown
rice our (about 3 g).
Constituent

a
b
c
d
e
f
g

Pretreatment method

Calibration

External validation

Math treatments

Scatter corrections

SECa

RSQb

SECVc

VRd

SEPe

Slope

R2f

SDg/SEP

Glutamic acid

0,0,1,1
0,0,1,1
1,4,4,1
1,8,8,1
2,4,4,1
2,8,8,1

None
SNV + D
SNV + D
SNV + D
SNV + D
SNV + D

0.063
0.065
0.064
0.067
0.057
0.058

0.943
0.938
0.941
0.936
0.953
0.949

0.067
0.067
0.067
0.069
0.065
0.063

0.936
0.933
0.935
0.932
0.939
0.941

0.006
0.009
0.001
0.002
0.000
0.002

0.068
0.070
0.067
0.071
0.069
0.066

1.002
0.993
1.009
1.006
0.964
1.003

0.924
0.918
0.925
0.917
0.921
0.927

3.632
3.529
3.687
3.479
3.580
3.742

Phenylalanine

0,0,1,1
0,0,1,1
1,4,4,1
1,8,8,1
2,4,4,1
2,8,8,1

None
SNV + D
SNV + D
SNV + D
SNV + D
SNV + D

0.015
0.014
0.013
0.014
0.013
0.013

0.957
0.959
0.967
0.962
0.969
0.967

0.016
0.016
0.015
0.015
0.015
0.014

0.947
0.950
0.958
0.954
0.956
0.960

0.002
0.000
0.002
0.002
0.000
0.001

0.017
0.017
0.017
0.017
0.017
0.016

1.035
1.007
1.002
1.012
0.992
1.020

0.939
0.931
0.939
0.934
0.935
0.946

3.941
3.941
3.941
3.941
3.941
4.188

Arginine

0,0,1,1
0,0,1,1
1,4,4,1
1,8,8,1
2,4,4,1
2,8,8,1

None
SNV + D
SNV + D
SNV + D
SNV + D
SNV + D

0.031
0.030
0.031
0.027
0.032
0.031

0.920
0.927
0.921
0.939
0.916
0.921

0.033
0.032
0.036
0.031
0.038
0.035

0.908
0.915
0.896
0.914
0.880
0.898

0.002
0.000
0.001
0.004
0.001
0.003

0.035
0.036
0.031
0.035
0.032
0.032

0.964
0.956
0.967
0.962
0.973
0.973

0.875
0.868
0.903
0.878
0.892
0.895

2.829
2.750
3.194
2.829
3.094
3.094

Bias

SEC: the standard error of calibration and the unit as %.

RSQ: coefcient of determination in calibration.
SECV: standard error of cross-validation and the unit as %.
1 VR: 1 minus the ratio of unexplained variance to total variance.
SEP: the standard error of prediction and the unit as %.
R2: coefcient of determination in external validation.
SD: standard deviation and the unit as %.

SEP: 0.0160.071%) with high SD/SEP ratios from 2.750 to 4.188.

Considering all parameters, especially the bias, slope, and SEP, R2
and the SD/SEP ratio in external validation, the prediction equations from the pretreatment with 2,8,8,1 and SNV + D were more
accuracy than the others. This pretreatment method was therefore
selected as the most suitable to be used for developing calibration
equations of the amino acid composition in brown rice our.
Based on the spectra of brown rice samples scanned with a full
cup, the effects of different pretreatments for calibration equations
(data not shown) were different from the spectra of brown rice our
(Table 2). The equations based on SNV + D scatter correction method were better than the ones without scatter correction. Equations
with mathematical treatment combined with scatter correction
showed better effect than those with only scatter correction. Furthermore, the best two pretreatments were 2,4,4,1 with SNV + D
and 2,8,8,1 with SNV + D because of their good results in both cross
validation and external validation. The latter was selected to develop the calibration equation for having relatively higher RSQ, R2,
1 VR, SD/SEP, and relatively lower bias, SEC, SECV, SEP even
though these two combined pretreatments were very similar. The
pretreatment of 2,8,8,1 with SNV + D of the two kinds of spectra
for brown rice grain and brown rice our was therefore suitable to
be used in calibration equations for these three amino acids.
3.3. Calibration equations with brown rice our
Based on above calibration effects of three amino acids, the pretreatment of 2,8,8,1 with SNV + D were consequently used to develop the calibration equations for all amino acids with brown
rice our. Table 3 shows the results obtained in the calibration
equations developed from samples scanned using the full cup,
i.e., about 3 g brown rice our. The equations for fourteen amino
acids (aspartic acid, threonine, serine, glutamic acid, glycine, alanine, valine, isoleucine, leucine, phenylalanine, lysine, arginine,
proline, and histidine) showed high coefcients of determination
for calibration (RSQ: 0.9050.967) and cross-validation (1 VR:
0.8940.960) and low standard errors for calibration and cross val-

idation. Low bias, standard errors of prediction and high coefcients of determination (R2: 0.8840.953) of the equations for
each individual amino acid, except for lysine (0.839) and proline
(0.837), were obtained in external validation. In addition, their
SD/SEP ratios ranged from 2.917 to 4.188, and the slope values of
the equations were close to 1. This implied that an accurate and
reliable estimation of these amino acids can be obtained by NIRS.
It also demonstrated that the simultaneous estimation of different
amino acids with calibration equations developed from a unique
calibration set is possible. Just as the equations for individual amino acid contents, the performance of the calibration equation for
the total of amino acids (TAA) was also accurate. But the performance of the calibration equation for tyrosine showing SEC, SECV,
SEP, RSQ, 1 VR, R2 at 0.025%, 0.028%, 0.033%, 0.836, 0.793 and
0.648 respectively was not very good. Similarly, the results obtained for methionine and cysteine were not satisfactory due to
very low values for the coefcients of determination in cross validation and external validation as well as the SD/SEP ratios (1.094
and 1.200, respectively), which makes these calibration equations
unsuitable for use in routine analysis.
Calibration equations developed on the basis of the sample
spectra with a medium-sized adapter (sample weight about
500 mg), showed similar accuracy and reliability as those developed with the full cup (Table 4). The developed equations for thirteen amino acids (aspartic acid, threonine, serine, glutamic acid,
glycine, alanine, valine, isoleucine, leucine, phenylalanine, arginine, proline, and histidine) had low SECV (0.0100.059%) and
SEP (0.0140.074%), and high 1 VR (0.8890.947), R2 (0.813
0.905) and SD/SEP (2.3573.333). The equation of TAA was also
successfully developed with low error values, high determination
coefcient and high SD/SEP. This implies that an accurate and reliable estimation of these amino acids can also be obtained with
NIRS from smaller samples (about 500 mg) although the performance of the calibration equations for lysine and tyrosine in smaller samples were not as good as that of the calibration equations
for larger samples (about 3 g), and the performance was poorer
for each amino acid in external validation.

279

B. Zhang et al. / Food Chemistry 127 (2011) 275281

Table 3
Calibration and external validation statistics in NIRS equations of brown rice our (about 3 g) with mathematical treatments (2,8,8,1) and scatter corrections (SNV + D).a
Constituent

Aspartic acid
Threonine
Serine
Glutamic acid
Glycine
Alanine
Cysteine
Valine
Methionine
Isoleucine
Leucine
Tyrosine
Phenylalanine
Lysine
Histidine
Arginine
Proline
TAA
a

Calibration

External validation

Range

SEC

RSQ

SECV

212
212
215
212
214
217
221
211
225
212
213
218
212
213
211
217
217
212

0.6571.307
0.2730.511
0.3400.700
1.3262.881
0.3410.643
0.4150.848
0.0840.215
0.4320.867
0.1330.285
0.2810.588
0.5591.177
0.2170.590
0.3580.786
0.2840.568
0.1690.363
0.5771.240
0.2850.581
6.90813.957

0.024
0.009
0.016
0.058
0.013
0.020
0.015
0.015
0.020
0.010
0.020
0.025
0.013
0.014
0.009
0.031
0.015
0.207

0.952
0.949
0.933
0.949
0.932
0.920
0.543
0.959
0.380
0.963
0.964
0.836
0.967
0.919
0.927
0.921
0.905
0.969

0.028
0.011
0.017
0.063
0.015
0.023
0.015
0.018
0.020
0.011
0.021
0.028
0.014
0.017
0.010
0.035
0.016
0.240

0.932
0.926
0.920
0.941
0.918
0.901
0.529
0.937
0.366
0.952
0.958
0.793
0.960
0.878
0.913
0.898
0.894
0.958

For abbreviations of SEC, RSQ, SECV, 1

VR

Bias

SEP

Slope

R2

SD/SEP

0.002
0.001
0.000
0.002
0.002
0.001
0.001
0.002
0.002
0.001
0.003
0.002
0.001
0.001
0.001
0.003
0.005
0.026

0.036
0.012
0.017
0.066
0.016
0.020
0.020
0.018
0.032
0.012
0.024
0.033
0.016
0.019
0.011
0.032
0.018
0.262

1.008
0.962
1.025
1.003
1.041
1.063
0.831
1.002
0.910
1.041
1.079
0.821
1.020
0.913
0.997
0.973
0.865
0.999

0.880
0.912
0.911
0.927
0.895
0.925
0.290
0.931
0.141
0.947
0.953
0.648
0.946
0.839
0.884
0.895
0.837
0.945

2.917
3.250
3.412
3.742
3.000
3.550
1.200
3.778
1.094
4.167
4.333
1.636
4.188
2.421
3.000
3.094
2.278
4.271

VR, SEP, R2 and SD see Table 2.

Table 4
Calibration and external validation statistics in NIRS equations of brown rice our (about 500 mg) with mathematical treatments (2,8,8,1) and scatter corrections (SNV + D)a.
Constituent

Aspartic acid
Threonine
Serine
Glutamic acid
Glycine
Alanine
Cysteine
Valine
Methionine
Isoleucine
Leucine
Tyrosine
Phenylalanine
Lysine
Histidine
Arginine
Proline
TAA
a

Calibration

External validation

Range

SEC

RSQ

SECV

210
213
211
211
214
215
223
211
221
211
211
215
211
218
211
213
215
209

0.6571.303
0.2690.512
0.3430.697
1.3262.870
0.3430.640
0.4170.845
0.0810.220
0.4360.862
0.1330.284
0.2840.584
0.5631.169
0.2200.585
0.3670.780
0.2770.571
0.1690.363
0.5931.223
0.2880.579
6.93613.942

0.027
0.011
0.014
0.055
0.014
0.019
0.015
0.017
0.019
0.011
0.021
0.027
0.014
0.017
0.009
0.029
0.015
0.193

0.937
0.929
0.941
0.954
0.924
0.928
0.608
0.946
0.423
0.956
0.959
0.804
0.961
0.877
0.915
0.924
0.902
0.973

0.028
0.011
0.015
0.059
0.015
0.021
0.015
0.018
0.020
0.012
0.023
0.028
0.015
0.020
0.010
0.032
0.016
0.207

0.931
0.923
0.932
0.947
0.915
0.916
0.559
0.937
0.351
0.945
0.950
0.785
0.954
0.839
0.899
0.906
0.889
0.969

For abbreviations of SEC, RSQ, SECV, 1

VR

Bias

SEP

Slope

R2

SD/SEP

0.004
0.000
0.000
0.004
0.002
0.003
0.000
0.002
0.001
0.003
0.004
0.001
0.003
0.002
0.002
0.004
0.005
0.049

0.044
0.015
0.019
0.074
0.020
0.026
0.019
0.023
0.028
0.015
0.032
0.027
0.021
0.024
0.014
0.038
0.017
0.336

0.935
0.949
0.972
0.954
0.938
0.989
0.903
0.952
1.179
1.004
1.008
0.895
0.980
0.901
0.974
0.940
0.850
0.953

0.826
0.848
0.890
0.909
0.820
0.860
0.401
0.880
0.304
0.905
0.904
0.739
0.904
0.731
0.813
0.853
0.837
0.911

2.386
2.600
3.053
3.338
2.400
2.731
1.263
2.957
1.250
3.333
3.250
2.000
3.190
1.917
2.357
2.605
2.412
3.330

VR, SEP, R2 and SD see Table 2.

3.4. Calibration equations with brown rice

The mathematical pretreatment method of 2,8,8,1 with scatter
corrections (SNV + D) was also used to develop the calibration
equation for brown rice. The results in Table 5 shows the cross-validation and external validation statistics for the calibration equations, with the samples (about 3 g) scanned with a full cup
adapter, developed for amino acid composition in brown rice. Calibration equations for valine, isoleucine, leucine, phenylalanine,
histidine and TAA showed a close relationship between NIRS and
reference values, with high coefcient of determination (1 VR:
0.8580.905) in cross-validation and (R2: 0.8530.890) in external
validation. For aspartic acid, threonine, serine, glutamic acid, glycine, alanine, lysine and arginine, the performances of the calibration equations were also good, with high coefcient of
determination (1 VR: 0.7510.897) in cross-validation and
slightly low coefcient of determination (R2: 0.7440.847) and
intermediate SD/SEP values (2.0002.375) in external validation.
But the performances of the calibration equation for tyrosine
(RSQ = 0.870; R2 = 0.743) and proline (RSQ = 0.774; R2 = 0.744),

were not very good since they had low SD/SEP values (1.929 and
1.952, respectively). Similarly, the results obtained for cysteine
and methionine were considerably poorer, with low values of coefcient of determination in cross validation (1 VR: 0.441 and
0.272) and external validation (R2: 0.287 and 0.210, respectively),
which makes these calibration equations unsuitable to use in routine analysis.
Cross-validation and external validation statistics for the calibration equations developed by using the same calibration set, in
which samples were scanned with a medium size adapter (sample
weight about 500 mg), is presented in Table 6. The performance of
the calibration equation for each amino acid, except for tyrosine
and lysine, was not as accurate as that/those with full cup; because
their coefcient of determination were relatively low both in crossvalidation and external validation. Surprisingly, with good coefcients of determination (R2: 0.865 and 0.843, respectively) and
low performance error (SEP: 0.038% and 0.027%, respectively),
the effects of calibration for tyrosine and lysine were similar to
those with full cup both in cross-validation and external validation.
In addition, the calibration equations for methionine or cysteine

280

B. Zhang et al. / Food Chemistry 127 (2011) 275281

Table 5
Calibration and external validation statistics in NIRS equations of brown rice (about 3 g) with mathematical treatments (2,8,8,1) and scatter corrections (SNV + D)a.
Constituent

Aspartic acid
Threonine
Serine
Glutamic acid
Glycine
Alanine
Cysteine
Valine
Methionine
Isoleucine
Leucine
Tyrosine
Phenylalanine
Lysine
Histidine
Arginine
Proline
TAA
a

Calibration

External validation

Range

SEC

RSQ

SECV

215
213
215
212
215
214
220
214
223
216
214
221
213
217
214
216
214
215

0.6371.323
0.2660.518
0.3350.707
1.2952.911
0.3380.647
0.4130.851
0.0820.218
0.4310.871
0.1310.287
0.2760.592
0.5471.188
0.2110.595
0.3610.784
0.2820.570
0.1660.366
0.5801.241
0.2860.582
6.79814.109

0.037
0.012
0.018
0.074
0.017
0.026
0.015
0.023
0.022
0.015
0.028
0.030
0.019
0.020
0.011
0.038
0.018
0.295

0.895
0.918
0.913
0.924
0.888
0.877
0.585
0.902
0.309
0.919
0.932
0.774
0.925
0.834
0.896
0.883
0.870
0.941

0.044
0.015
0.022
0.087
0.020
0.029
0.017
0.027
0.022
0.018
0.036
0.036
0.024
0.024
0.013
0.044
0.020
0.376

0.852
0.879
0.875
0.897
0.850
0.841
0.441
0.862
0.272
0.879
0.887
0.695
0.887
0.751
0.858
0.837
0.838
0.905

For abbreviations of SEC, RSQ, SECV, 1

VR

Bias

SEP

Slope

R2

SD/SEP

0.002
0.001
0.003
0.010
0.000
0.002
0.001
0.001
0.001
0.000
0.001
0.001
0.001
0.000
0.002
0.001
0.003
0.016

0.049
0.017
0.026
0.104
0.022
0.031
0.020
0.027
0.030
0.017
0.040
0.028
0.023
0.023
0.013
0.042
0.021
0.411

0.859
0.873
0.864
0.858
0.881
0.958
0.787
0.918
1.044
0.922
0.886
0.813
0.912
0.941
0.940
0.890
0.827
0.871

0.808
0.828
0.827
0.847
0.807
0.811
0.287
0.853
0.210
0.888
0.868
0.743
0.890
0.744
0.840
0.837
0.780
0.885

2.143
2.294
2.231
2.375
2.182
2.290
1.200
2.519
1.167
2.941
2.600
1.929
2.913
2.000
2.538
2.357
1.952
2.723

VR, SEP, R2 and SD see Table 2.

Table 6
Calibration and external validation statistics in NIRS equations of brown rice (about 500 mg) with mathematical treatments (2,8,8,1) and scatter corrections (SNV + D)a.
Constituent

Aspartic acid
Threonine
Serine
Glutamic acid
Glycine
Alanine
Cysteine
Valine
Methionine
Isoleucine
Leucine
Tyrosine
Phenylalanine
Lysine
Histidine
Arginine
Proline
TAA
a

Calibration

External validation

Range

SEC

RSQ

SECV

212
213
210
209
210
212
218
210
217
210
207
213
209
214
209
210
211
209

0.6351.324
0.2620.518
0.3300.709
1.2902.910
0.3360.647
0.4070.854
0.0810.219
0.4260.872
0.1300.287
0.2730.595
0.5471.184
0.2100.595
0.3570.787
0.2780.572
0.1640.367
0.5691.243
0.2860.579
6.77914.110

0.038
0.015
0.019
0.075
0.016
0.025
0.015
0.022
0.020
0.014
0.027
0.029
0.020
0.022
0.011
0.037
0.022
0.310

0.893
0.880
0.912
0.924
0.904
0.883
0.576
0.911
0.442
0.929
0.933
0.803
0.925
0.804
0.897
0.892
0.789
0.936

0.048
0.019
0.023
0.093
0.021
0.032
0.017
0.028
0.021
0.018
0.035
0.034
0.027
0.026
0.014
0.044
0.025
0.391

0.825
0.800
0.868
0.883
0.838
0.812
0.470
0.860
0.330
0.889
0.891
0.732
0.864
0.715
0.841
0.850
0.734
0.898

For abbreviations of SEC, RSQ, SECV, 1

VR

Bias

SEP

Slope

R2

SD/SEP

0.005
0.004
0.004
0.023
0.000
0.002
0.001
0.001
0.001
0.003
0.002
0.001
0.002
0.003
0.003
0.000
0.001
0.002

0.048
0.019
0.025
0.107
0.022
0.031
0.020
0.020
0.032
0.031
0.020
0.038
0.032
0.027
0.024
0.015
0.046
0.456

0.958
0.961
0.965
0.955
0.954
1.007
0.968
0.968
0.917
0.829
0.991
1.038
0.843
1.022
0.980
1.038
0.910
0.924

0.787
0.758
0.819
0.814
0.788
0.802
0.314
0.314
0.790
0.154
0.835
0.865
0.638
0.843
0.723
0.802
0.782
0.707

2.188
2.053
2.320
2.308
2.182
2.290
1.200
3.400
1.094
1.613
5.200
1.421
2.094
1.704
1.375
6.600
0.891
2.454

VR, SEP, R2 and SD see Table 2.

were not suitable to use in routine analysis because of the low

coefcient of determination and the large error in external
validation.
4. Discussion
NIRS has been successfully correlated with the traditional
methods of quality analysis that have been commonly used for
amino acids. The statistics of NIRS calibration and external validation for estimation of 15 different amino acids, except methionine
and cysteine (which gave quite low coefcients of determination in
the calibration and validation), are satisfactory and useful for
screening in brown rice. The large variance in the reference data
is good for developing calibration equations and the inclusion of
a wide range of sample varieties can make the calibration equations to possess a powerful predictive ability (Velasco & Becker,
1998). The good results obtained for these amino acids can be explained on the basis of their large variation in the calibration set
and having high potential to be used for analytical application.
The calibration results in the present project were similar to those
on milled rice (Wu et al., 2002). The sample status of intact rice

grain, milled rice or milled rice our had been employed in most
studies on developing NIRS calibration models for rice quality
characteristics. Wu et al. (2002) had compared the calibrations of
the amylose content, gel consistency and alkali spreading value
with different rice categories, including brown rice, brown rice
our, milled rice and milled rice our, which demonstrated that
the models developed using brown rice and milled rice were nearly
equal or superior to those using intact rice grains, but slightly
poorer to those using their corresponding our samples. In this
study, a similar situation for amino acids was observed. Calibration
equations based on brown rice our samples were more accurate
than those based on brown rice samples. Furthermore, the effect
of the calibrations based on different sample mass has been compared, which indicated that the calibration equations for large
samples give slightly better results than those for small samples,
which might be explained by diversity of the surface area and size
of the seeds used (Pazdernik et al., 1997).
Traditionally, rice was milled into white rice as major source of
nutrients for humans. According to nutritionists, brown rice should
be more preferred because of its higher nutritional value. In spite
of the elevated content of nutritional and bio-functional compo-

B. Zhang et al. / Food Chemistry 127 (2011) 275281

nents, brown rice is not considered suitable for the dining table because of its dark appearance, hard texture, and poor cooking quality. Recently rice consumers have been requesting an improvement
in of the nutritional and industrial values for this staple food. Consequently, optimizing milling procedures with minimum breakage; retaining the maximum possible nutrition of brown rice and
with preferable cooking attributes, has been the primary goal of
rice processing industries (Das, Gupta, Kapoor, Banerjee, & Bal,
2008; Liang et al., 2008). On the other hand, with people appetency
for more healthy food, a great amount of nutritional and bio-functional foodstuff made from brown rice will be demanded. Recently,
a new processing technology for improving the taste of brown rice,
was applied to the production of novel foodstuff from the extruded
pre-germinated brown rice, which contained more nutritional and
bio-functional components than ordinary rice products. Consequently, increasing the content of essential amino acids, especially
for lysine, cysteine and tryptophan which are limiting amino acids
in rice, is very important for improving nutritional quality in rice
breeding. Furthermore, improved methodologies for screening
intermediate lines more efciently and economically are needed
by the breeders. Single plant analysis or mass analyses of samples
from different cultivars with NIRS can be employed in many research areas such as the intermediate material selection in a breeding project, special quality plants screening in germplasm,
mutation library development, and genetic material analyses in
big populations (Wu et al., 2002). The results of this experiment
showed that the NIRS calibrations for amino acids can be directly
used in breeding programs and classication of brown rice in the
food industry.
Although good calibration equations were obtained for most
amino acids, the equations for methionine and cysteine did not
show satisfactory coefcients of determination. The reason why
there was difculty in developing successful equations for these
amino acids is possibly explained by the chemical determination
method. By using HPLC method based HCl hydrolysis parts of
methionine and cysteine were destroyed during the sample processing resulting in their reference values having insufcient accuracy for NIRS calibration. Other alkali hydrolysis methods
maintaining these two amino acids could be applied if reliable calibration equations need to be developed for them. Moreover, tryptophan, as a particularly limiting amino acid in rice, which was
destroyed completely by HCl hydrolysis (Sotelo, Hernandez, Montalvo, & Sousa, 1994), was not detected by the method used here.
New methods, such as the uorometric method, could be valuable
in developing NIRS calibration of tryptophan in future studies.
Acknowledgements
This work was supported by the projects (No. 2010C32002, No.
2007C22016 and No. 2007C12902) of Science and Technology Ofce of Zhejiang Province by the Zhejiang Provincial Natural Science
Foundation of China (No. Y3080217) and by the 151 Program for
the Talents of Zhejiang Province. We thank Mr. Yuanlong Li, the
technician of 985-Institute of Agro-biology and Environmental Sciences of Zhejiang University, for the assistance in using the L-8900
Amino Acid Analyzer. English improvement by Mr. Quampah Alfre
Djulius was appreciated.
References
Bao, J. S., Cai, Y. Z., & Corke, H. (2001). Prediction of rice starch quality parameters by
near-infrared reectance spectroscopy. Journal of Food Science, 66, 936939.
Bao, J. S., Shen, Y., & Jin, L. (2007). Determination of thermal and retrogradation
properties of rice starch using near-infrared spectroscopy. Journal of Cereal
Science, 46, 7581.

281

Barnes, R. J., Dhanoa, M. S., & Lister, S. J. (1989). Standard normal variate
transformation and detrending for near infrared diffuse reectance spectra.
Applied Spectroscopy, 43, 772777.
Das, M., Gupta, S., Kapoor, V., Banerjee, R., & Bal, S. (2008). Enzymatic polishing of
rice A new processing technology. Food Science and Technology, 41,
20792084.
Delwiche, S. R., McKenzie, K. S., & Webb, B. D. (1996). Quality characteristics in rice
by near-infrared reectance analysis of whole-grain milled samples. Cereal
Chemistry, 73, 257263.
Fontaine, J., Hrr, J., & Schirmer, B. (2001). Near infrared spectroscopy enables the
fast and accurate prediction of the essential amino acid contents in soy,
rapeseed meal, sunower meal, peas, shmeal, meat meal products, and poultry
meal. Journal of Agriculture and Food Chemistry, 49, 5766.
Gonzlez-Martn, I., lvarez-Garca, N., & Gonzlez-Cabrera, J. M. (2006). Nearinfrared spectroscopy (NIRS) with a bre-optic probe for the prediction of the
amino acid composition in animal feeds. Talanta, 69, 706710.
Hruschka, W. R. (1987). Data analysis: Wavelength selection methods. In P.
Williams & K. Norris (Eds.), Near-infrared technology in the agricultural and
food industries. St. Paul: American Association of Cereal Chemists (pp. 35
55).
Itani, T., Tamaki, M., Arai, E., & Horino, T. (2002). Distribution of amylose, nitrogen,
and minerals in rice kernels with various characters. Journal of Agricultural and
Food Chemistry, 50, 53265332.
Lamberts, L., Bie, E. D., Vandeputte, G. E., Veraverbeke, W. S., Derycke, V., Man, W. D.,
et al. (2007). Effect of milling on color and nutritional properties of rice. Food
Chemistry, 100, 14961503.
Li, J. X., Ming, S. G., Zhang, H. L., Yan, L. L., Luo, C. B., & Li, Z. C. (2006). The PLS
calibration model optimization and determination of rice protein content by
near-infrared reectance spectroscopy. Spectroscopy and Spectral Analysis, 26(5),
833837.
Li, W. S., & Shaw, J. T. (1997). Determining the fat acidity of rough rice in near
infrared reectance spectroscopy. Cereal Chemistry, 74(5), 556560.
Liang, J. F., Li, Z. G., Tsuji, K., Nakano, K., Robert, N. M. J., & Hamer, R. J. (2008). Milling
characteristics and distribution of phytic acid and zinc in long-, medium- and
short-grain rice. Journal of Cereal Science, 48, 8391.
Lu, G. Q., Huang, H. H., & Zhang, D. P. (2006). Prediction of sweetpotato starch
physiochemical quality and pasting properties using near-infrared reectance
spectroscopy. Food Chemistry, 94, 632639.
Mitchell, H. H., & Block, R. J. (1946). Some relationships between the amino acid
content of proteins and their nutritive values for the rat. The Journal of Biological
Chemistry, 163(3), 599620.
Ohtsubo, K., Suzuki, K., Yasui, Y., & Kasumi, T. (2005). Bio-functional components in
the processed pre-germinated brown rice by a twin-screw extruder. Journal of
Food Composition and Analysis, 18, 303316.
Pazdernik, D. L., Killam, A. S., & Orf, J. H. (1997). Analysis of amino acid and fatty acid
composition in soybean seed, using near-infrared reectance spectroscopy.
Agronomy Journal, 89, 679685.
Shenk, J. S., & Westerhaus, M. O. (1993). Analysis of agriculture and food products by
near infrared reectance spectroscopy. Monograph (pp. 4451, Vol. 26). Port
Matilda, PA: Infrasoft International.
Shenk, J. S., & Westerhaus, M. O. (1996). Calibration the ISI way. In A. M. C. Davies, &
P. C. Williams (Eds.), Near infrared spectroscopy: The future waves (pp. 198202).
Chichester, UK: NIR Publications.
Shenk, J. S., & Westerhaus, M. O. (1991a). Population denition, sample selection,
and calibration procedures for near infrared spectroscopy. Crop Science, 31,
469474.
Shenk, J. S., & Westerhaus, M. O. (1991b). Population structuring of near infrared
spectra and modied partial least squares regression. Crop Science, 31,
15481555.
Shu, Q. Y., Wu, D. X., Xia, Y. W., Gao, M. W., & McClung, A. (1999). Analysis of grain
quality characters in small ground brown rice samples by near infrared
reectance spectroscopy. Scientica Agronomica Sinica, 32(4), 9297.
Sotelo, A., Hernandez, M., Montalvo, I., & Sousa, V. (1994). Amino acid content and
protein biological evaluation of 12 Mexican varieties of rice. Nutrition, 71,
605609.
Velasco, L., & Becker, H. C. (1998). Estimating the fatty acid composition of the oil in
intact-seed rapeseed (Brassica napus L.) by near-infrared reectance
spectroscopy. Euphytica, 101, 221230.
Wu, J. G., & Shi, C. H. (2003). The applications of near-infrared reectance
spectroscopy in plant breeding and germplasm research. Journal of Plant
Genetic Resources, 4(1), 6872.
Wu, J. G., & Shi, C. H. (2004). Prediction of grain weight, brown rice weight and
amylose content in single seeds of rice using near-infrared reectance
spectroscopy. Field Crops Research, 87, 1321.
Wu, J. G., & Shi, C. H. (2007). Calibration model optimization for rice cooking
characteristics by near infrared reectance spectroscopy (NIRS). Food Chemistry,
103, 10541061.
Wu, J. G., Shi, C. H., & Zhang, X. M. (2002). Estimating the amino acid composition in
the milled rice powder by near-infrared reectance spectroscopy. Field Crops
Research, 75(1), 17.
Xie, X. H., Xiao, X., Li, X. F., & Li, Y. R. (2006). Mathematic models for
analyzing protein content in single brown rice with near infrared
spectroscopy. Transactions of the Chinese Society for Agricultural Machinery,
37(8), 120122.