Professional Documents
Culture Documents
a r t i c l e
i n f o
Article history:
Received 21 April 2015
Received in revised form 23 July 2015
Accepted 4 August 2015
Available online 10 August 2015
Keywords:
Molecular uorescence
Second-order chemometric methods
Optical brighteners
Laundry powder
Environmental samples
a b s t r a c t
This study describes a novel approach for the simultaneous determination of CBS-X and CXT as widely used optical brighteners in household detergent, by combining the advantage of the high sensitivity of molecular uorescence, and the selectivity of second-order chemometric methods. The proposed method is assisted by secondorder chemometric analyses employing the PARAFAC, SWATLD and APTLD that help us to determine CBS-X
and CXT in laundry powders and environmental samples, through the unique decomposition of the three-way
data array. Proposed method can provide the extraction of relative concentrations of the analytes, as well as
the spectral proles. This approach achieves the second-order advantage and in principle could be able to overcome the spectral uncalibrated interference problems in the determination of CBS-X and CXT at the ng g1 level.
By spiking the known concentrations of these compounds to the real samples, the accuracy of the proposed
methods was validated and recoveries of the spiked values were calculated. High recoveries (90.00%113.33%)
for the spiked laundry powders and real environmental samples indicate the present method successfully
faces this complex challenge without the necessity of applying separation and preconcentration steps in environmental contaminations.
2015 Elsevier B.V. All rights reserved.
1. Introduction
Nowadays, variety types of laundry detergents such as laundry powder, laundry soaps and liquid laundry are produced in the whole of the
world. Laundry detergents contain several chemical components, including surfactant, optical brightener, bleaching agent and builder.
Optical brighteners (OBs) or uorescent whitening agents (FWAs in
the detergent industry) are synthetic chemical compounds that stick to
the surface of the clothing or fabrics. These particles make fabrics seem
cleaner by absorbing ultraviolet light (usually 340370 nm), and reemitting it as blue light (typically 420470 nm), thus they enhance
the intensity of visual reection and strength the optical impression of
whiteness and brightness. Fluorescent whitening agents that are used
in detergents are moderately water-soluble organic compounds with a
high afnity for cellulosic material. When they adsorb to the textiles,
their strong blue uorescence eliminate the yellowish cast of white fabrics and make them look whiter [1].
Optical brighteners can be classied, based on their structure and
properties. A number of uorescent compounds such as quinolone, coumarin, benzoxazole, and distyrylbiphenyl are used as uorescent whitening agents. All OBs are highly substituted ring (aromatic) structures
Corresponding author.
E-mail address: agholami@kashanu.ac.ir (A. Gholami).
http://dx.doi.org/10.1016/j.saa.2015.08.012
1386-1425/ 2015 Elsevier B.V. All rights reserved.
and contain many double bonds that can be activated by UV light [2].
These components are used in textiles, papers, plastics and synthetic bers, plus widespread use in some medical, chemical, cosmetics and petroleum applications. Around 80% of FWAs that are used in paper, textile
and detergent formulations are based on the stilbene derivatives.
Benzenesulfonic acid, 2,2([1,1-biphenyl]-4,4-diyldi-2,1-ethenediyl)
bis-disodium salt or Tinopal CBS-X (Fig. 1-A) and disodium 4,4-bis[(4anilino-6-morpholion-1,3,5-trizin-2-yl)amino]stilbene-2,2-disulphonate
or CXT (Fig. 1-B) are most widely used in detergent industries [3].
CBS-X is used in household detergents at concentration levels
between 0.05% and 0.15% and undergoes photo degradation N50% in 12
months. CXT is used in detergents at concentration levels between 0.02%
and 0.10% and undergoes photo degradation N 70% in 28 days [4,5].
Residues of OBs are left on our clothes, and they can cause allergic reactions when absorbed by skin. Detergent industries due to the ubiquitous use of optical brightener are important potential sources for the
pollution of natural waters and ecosystem. Laundry wastewater is the
largest supplier of optical brighteners to wastewater systems. Therefore,
detection of optical brighteners in aquatic systems can help water municipalities and avoid increased anthropogenic input that may greatly
affect ecosystems. When OBs enter in the water system, they cause a potential hazard to aquatic life. This pollution remains in wastewater for
long periods and negatively affects water quality, animal and plant
life. Brighteners are highly susceptible to adsorption. They are removed
A. Gholami et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 153 (2016) 108117
109
Fig. 1. (a) Structure of CBS-X , (b) structure of CXT (c) Excitation spectrum of CBS-X and CXT and (d) Emission spectrum of CBS-X and CXT.
from surface and ground water by adsorption into the soil and tend to
accumulate on sewage sludge. This allows researchers to assess the effectiveness of natural cleansing of wastewater by determining brightener concentrations in ground or surface waters [6].
The increasing public health concerns have encouraged us to consider the content and distribution of FWAs in laundry detergent as a quality
control parameter. The amount of FWAs in textile and paper products is
usually determined by direct spectrophotometric method [711]. These
methods do not give individual concentration of FWAs and measure
only total FWA concentrations. Environmental samples usually contain
a number of organic species at relatively high concentrations, which can
interfere with the optical brightener's quantication.
Since 1976, high performance liquid chromatography (HPLC) has been
applied to separate and determine FWAs in detergents and environmental
samples [1218]. Reversed-phase HPLC with ion-pair chromatography
110
A. Gholami et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 153 (2016) 108117
Analysis of complex matrices into their pure-component contributions with unknown components without separation is a fundamental
problem in analytical chemistry. Spectrum overlapping was resolved
by multivariate quantication methods such as classical least squares
(CLS), inverse least squares (ILS) and partial least square (PLS). PLS and
related chemometric methods need a large number of calibration and
test sets and also for providing a robust regression model, all analytes
and interferences have to be in the calibration set at the appropriate concentration level. Analysis and quantication of analytes in complex matrices are achievable without any previous knowledge about their
chemical substances by chemometric methods. The philosophy behind
second order or higher order data analysis is quantication of analytes
even in the presence of unmodeled or uncalibrated interferences and
this property, in fact, is known as second order advantage [24].
Environmental samples contain different organic materials such as
surfactants, chromophoric compounds, synthetic colors and poly hydrocarbon aromatic (PAH) which have been intense overlapping with the
analytes.
Unfortunately, PLS and related chemometric methods do not exhibit
the second order advantage, therefore all components in the complex
samples must be known, but identication of all components in complex matrices is impossible and time consuming.
Second order multivariate calibration methods were successfully applied for three-way array of data that is obtained by stacking the data
matrices of different samples under each other.
Second-order data is usually brought from hyphenated instruments,
such as an excitationemission matrix (EEM) of spectrouorometer and
high-performance liquid chromatography with a diode array detector
(HPLC-DAD), gas chromatography with a mass detector (GC-MS), and
absorbance spectra-pH data [25,26].
Aim of this study is applying uorescence spectroscopy to routinely
determine FWAs in laundry detergent, wastewater and agricultural soil
samples. Fluorescence spectroscopy as an analytical technique has the
advantage of selectivity and sensitivity in comparison with other spectroscopic methods.
An improved way to analyze the data is exciting the sample at different excitation wavelengths to obtain the emission spectra. Fluorescence
data array, X, can be produced by stacking a series of excitationemission
spectra for each sample to create an excitation emission matrix (EEM).
Each EEM is a matrix and by incorporating several samples, a three-way
array (sample emission excitation) is obtained as a landscape that
covers the total area of uorescence. This three-way array can be processed by any of the second-order algorithms such as parallel factor analysis (PARAFAC), alternating penalty trilinear decomposition (APTLD) and
self-weighted alternating trilinear decomposition (SWATLD). The presentation of data in an EEM format is a good way of visualizing the different uorophores in the sample. But apart from the possibility of
characterizing distinct peaks, it is difcult to unravel the individual
uorophores in the sample [27]. The present work attempts to demonstrate the combination of EEM array in a photoluminescence system
assisted by three-way calibration method for simultaneous determination of optical brighteners with strongly overlapping spectra in the presence of uncalibrated interferences. The proposed method is the rst
application of multi-way methods in the multi-dimensional analysis of
the optical brighteners in laundry detergent and environmental samples.
2. Theory
2.1. PARAFAC modeling
PARAFAC is usually employed for modeling trilinear data arrays. The
PARAFAC model can be written as following equation.
where xijk is the intensity of the ith sample at the jth variable (emission
mode) and at the kth variable (excitation mode), F is the total number of
responsive components, aif is the relative concentration of component f
in the ith sample, bjf is the signal intensities at emission wavelength j
and ckf is the signal intensities at excitation wavelength k in each dimension for component f.
Stacking a series of sample concentration matrices can make up a
data cube X; the algorithm decomposes the cube X into three matrices
A, B and C with elements aif, bjf, ckf, respectively. The algorithm proceeds
to minimize the sum of squares of the residuals, eijk, in the model using
alternating lease squares (ALS) until convergence is reached [28].
2.2. APTLD modeling
Alternating penalty trilinear decomposition (APTLD) is applied for
the decomposition of three-way data arrays. Performance of this algorithm is based on utilizing alternating penalty factors which minimizes
three new least squares-based constrained objective functions. The
values of penalty factors p, q and r should be chosen before implementation of the algorithm. In additions, this method is insensitive to increasing number of component by choosing a large number of penalty
factors, thus avoiding the complexity of determining a correct component number for the model which is intrinsic in the PARAFAC algorithm.
It has the second-order advantage for determination of analytes in the
presence of unknown interference. PARAFAC can give degenerate solutions when two components of the same mode have the signals with
high correlations. This problem called two-factor degeneracy. The
APTLD algorithm can avoid the two-factor degeneracy problem and
the proposed algorithm can overcome the slow convergence, which is
difcult to handle for the traditional PARAFAC algorithm.
The common algorithm for the APTLD method is as follows:
1. Obtain the accurate number of species.
2. Randomly initialize B and C by choosing suitable penalty factors p, q
and r.
3. Compute A from B and C using least-squares regression.
4. B and C were computed in the same way as matrix A in step 3.
5. Provide column wise normalization by scaling B and C.
6. Repeat steps 24 until a stopping criterion is satised [29,30].
2.3. SWATLD modeling
A self-weighted alternating trilinear decomposition (SWATLD) derived from the alternating trilinear decomposition (ATLD) algorithm.
The SWATLD algorithm is based on a least-squares scheme, in which diagram operation is added to each loss function, making it more stable and
exible, in other words, an overestimation of the number of factors will
not affect on meaningful obtained results by SWATLD. There are different
features for SWATLD such as decreasing the computation time in each iteration for optimizing procedure and being insensitive to the excess estimated factors in the model, the unique optimizing approach and
quantication of analytes even in the presence of unknown interferences.
This algorithm is based on the following procedure:
1.
2.
3.
4.
5.
3. Experimental section
3.1. Materials
xi jk f 1 if bjf ck f ei jk
A. Gholami et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 153 (2016) 108117
111
112
A. Gholami et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 153 (2016) 108117
Laundry powders
Samples 1 and 2
Sample 3
Agriculture soils
Samples 1
Sample 2
Wastewater
Samples 1
Number of
components
SD of
residual
CV
1
2
3
4
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5
6
1
2
3
4
5
6
7
3.2
2.3
1.7
1.7
3.7
2.6
1.8
1.3
1.3
7.8
5.4
2.4
1.7
1.7
16.0
8.0
5.0
2.3
1.5
1.4
13
8.6
5.9
2.8
2.3
2.0
2.0
200.3
100.1
50.2
49.2
20.2
10.5
6.3
1.2
1.2
49.8
12.5
8.1
2.1
2.1
45.5
12.3
9.2
5.8
1.1
1.1
30.0
11.1
8.7
8.5
5.4
1.4
1.4
Laundry powders
Sample 1
Sample 2
Sample 3
Agriculture soils
Sample 1
Sample 2
Wastewater
Sample 1
Methods
CBS-X%
CXT%
PARAFAC
APTLD
SWATLD
PARAFAC
APTLD
SWATLD
PARAFAC
APTLD
SWATLD
PARAFAC
APTLD
SWATLD
PARAFAC
APTLD
SWATLD
PARAFAC
APTLD
SWATLD
5.24 102
5.37 102
5.37 102
0.00
0.00
0.00
5.02 102
5.10 102
5.10 102
1.31 104
1.21 104
1.21 104
3.31 105
3.40 105
3.40 105
NA
1.08 105
1.08 105
1.13 101
1.14 101
1.14 101
1.09 101
1.02 101
1.02 101
0.00
0.00
0.00
2.78 104
2.86 104
2.86 104
7.88 105
5.73 105
5.73 105
NA
3.07 105
3.07 105
A. Gholami et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 153 (2016) 108117
113
Fig. 2. Obtained emission (A) and excitation (B) loadings from PARAFAC algorithm for laundry powder sample 3.
Table 3
Obtained recoveries of analytes in laundry powder samples.
Sample 1
Methods
Added CBS
(ng g1)
Found CBS
(ng g1)
Expected
CBS
(ng g1)
Recovery (%)
Added CXT
(ng g1)
Found CXT
(ng g1)
Expected
CXT
(ng g1)
Recovery (%)
PARAFAC
0.00
1.00
2.00
0.00
1.00
2.00
0.00
1.00
2.00
0.00
1.50
3.00
0.00
1.5
3.00
0.00
1.5
3.00
0.00
1.0
2.0
0.0
1.0
2.0
0.00
1.0
2.0
5.96
6.64
7.88
6.11
6.86
7.69
6.11
6.86
7.69
1.70
3.22
1.37
2.89
1.37
2.89
5.71
6.89
7.05
5.80
6.83
7.81
5.80
6.83
7.81
6.96
7.96
7.11
8.11
7.11
8.11
1.50
3.00
1.50
3.00
1.50
3.00
6.71
7.71
6.80
7.80
6.80
7.80
95.40
98.99
96.48
94.82
96.48
94.82
113.33
107.33
91.33
96.34
91.33
96.34
102.68
91.43
100.44
100.12
100.44
100.12
0.00
5.00
10.00
0.00
5.00
10.00
0.00
5.00
10.00
0.00
5.00
10.00
0.00
5.00
10.00
0.00
5.0
10.0
0.00
10.0
15.0
0.00
10.0
15.0
0.00
10.0
15.0
12.85
17.45
24.16
12.97
17.59
24.53
12.97
17.59
24.53
24.83
28.83
34.62
23.21
30.00
36.50
23.21
30.00
36.50
17.85
22.85
17.97
22.97
17.97
22.97
29.83
34.83
28.21
33.21
28.21
33.21
10.0
15.0
10.0
15.0
10.0
15.0
97.75
105.73
97.88
106.79
97.88
106.79
96.64
99.39
106.34
109.90
106.34
109.90
90.00
92.26
108.00
99.46
108.00
99.46
APTLD
SWATLD
Sample 2
PARAFAC
APTLD
SWATLD
Sample 3
PARAFAC
APTLD
SWATLD
9.00
13.84
10.80
14.92
0.0
10.80
14.92
114
A. Gholami et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 153 (2016) 108117
Fig. 3. Emission (A) and excitation (B) loadings that are provided by the ve-component PARAFAC model for agricultural soil sample 2.
tting model or DTLD in all samples can repeatedly give almost identical
results for three algorithms in all samples. Because a best-tting model
needs more iteration setups and consumes an extended computation
time, DTLD is used as an initial value estimation that provides less iteration. By using DTLD as an initial value and choosing a suitable number
of component, the PARAFAC, SWATLD and APTLD can provide unique
results (Table S3).
Number of iterations and recoveries were calculated and listed in
Table S4 by using four and ve as component numbers with DTLD as
an initial value for one of the laundry powder samples. The average of
the iteration number for APTLD (IT = 45) is same as SWATLD and less
than PARAFAC (IT = 210). The results disclose that the APTLD and
SWATLD algorithms not only resolve the proles accurately but also
overcome the slow convergence. Comparison between proposed
methods reveals that any clear changes were not observed in SWATLD
and APTLD by increasing number of components in contrast to
PARAFAC. SWATLD and APTLD are insensitive to the estimated component number, therefore avoiding the difculty of determining a correct
component number for the models, which is an initial and important
stage in the PARAFAC algorithm model.
The degree of overlapping between the proles of all components in
the data matrix is appropriate so these three algorithms can provide a
same reasonable result if a sufcient number of components be selected. Three algorithms can predict the reasonable result and have good
agreement with each other for laundry powder and agricultural soil
samples.
A. Gholami et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 153 (2016) 108117
115
Fig. 4. Obtained emission (A) and excitation (B) loadings from SWATLD for wastewater sample.
In this formulation, yi and i are the loadings for the given analyte
assessed from the developed algorithms and the loadings obtained
from the calibration line, respectively, and i is the number of used calibration samples.
Limit of detection (LOD) was calculated directly from the calibration
curve [40]. LOD was calculated as 3.3/SEN, where is the estimation of
random error in the y direction and equal to:
s
XI
^ i 2
y y
i1 i
I2
116
A. Gholami et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 153 (2016) 108117
Table 4
Recoveries of CBS-X and CXT mixtures in environmental samples.
Agriculture soil 1
Methods
Added CBS
(ng g1)
Found CBS
(ng g1)
Expected CBS
(ng g1)
Recovery (%)
Added CXT
(ng g1)
Found CXT
(ng g1)
Expected CXT
(ng g1)
Recovery (%)
PARAFAC
0.00
1.00
2.00
0.00
1.00
2.00
0.00
1.00
2.00
0.00
1.00
2.00
0.00
1.00
2.00
0.00
1.00
2.00
0.00
1.00
2.00
0.00
1.00
2.00
0.00
1.00
2.00
5.99
6.82
7.40
5.54
6.61
7.43
5.54
6.61
7.43
3.77
4.46
5.42
3.87
4.67
5.63
3.87
4.67
5.63
5.06
4.29
4.25
4.92
5.61
6.98
4.92
5.61
6.98
6.99
7.99
6.54
7.54
6.54
7.54
4.77
5.77
4.87
5.87
4.87
5.87
6.06
7.06
5.92
6.92
5.92
6.92
97.56
92.61
101.08
98.54
101.08
98.54
93.50
93.93
95.89
95.91
95.89
95.91
71.00
60.22
94.76
95.08
94.76
95.08
0.00
5.00
10.00
0.00
5.00
10.00
0.00
5.00
10.00
0.00
5.00
10.00
0.00
5.00
10.0
0.00
5.00
10.0
0.00
5.00
10.00
0.00
5.00
10.00
0.00
5.00
10.00
12.68
16.98
22.43
13.03
18.68
25.12
13.03
18.68
25.12
8.96
13.57
17.74
6.52
12.31
17.67
6.52
12.31
17.67
0.0
13.87
33.39
13.98
18.20
24.20
13.98
18.20
24.20
17.68
22.68
18.03
23.03
18.03
23.03
14.96
18.96
11.52
16.52
11.52
16.52
5.00
10.00
18.98
23.98
18.98
23.98
96.04
98.89
103.60
109.07
103.60
109.07
90.70
93.56
106.85
106.96
106.85
106.96
275.45
333.90
95.89
100.91
95.89
100.91
APTLD
SWATLD
Agriculture soil 2
PARAFAC
APTLD
SWATLD
Wastewater
PARAFAC
APTLD
SWATLD
the clean-up column (with appropriate pH) and eluted the column with
a suitable solvent and nally analysis could be performed by HPLC. This
chromatography method is imprecise, laborious, need to toxic and expensive pure organic solvent.
As a result, there is much interest to improve the uorometric method that requires inexpensive and simple-to-operate equipment for analysis of OBs in real samples. One of the major shortcomings of
uorometric analysis of OBs is interferences from natural organic
Table 5
Figures of merit obtained from the multivariate calibration procedures in real samples.
Samples
Laundry powders
PARAFAC
APTLD
SWATLD
Agriculture soils
PARAFAC
APTLD
SWATLD
Wastewater
PARAFAC
APTLD
SWATLD
Sensitivity
Precision
Detection limit
Sensitivity
Precision
Detection limit
Sensitivity
Precision
Detection limit
Sensitivity
Precision
Detection limit
Sensitivity
Precision
Detection limit
Sensitivity
Precision
Detection limit
Sensitivity
Precision
Detection limit
Sensitivity
Precision
Detection limit
Sensitivity
Precision
Detection limit
CBS-X
CXT
792.6
0.22
0.72
669.2
0.22
0.73
669.2
0.22
0.73
650
0.28
0.92
651.8
0.39
1.28
651.8
0.39
1.28
532.69
0.46
1.53
532.69
0.46
1.53
120
1.37
4.53
133.2
1.5
4.94
133.2
1.5
4.94
104.5
1.69
5.58
105
1.66
5.48
105
1.66
5.48
97.8
1.33
4.40
97.8
1.33
4.40
A. Gholami et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 153 (2016) 108117
Acknowledgments
The authors are grateful to the University of Kashan for supporting
this work by grant no. 363018/2.
Appendix A. Supplementary data
Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.saa.2015.08.012.
References
[1] J.A. Field, T.M. Field, T. Poiger, H. Siegrist, W. Giger, Water Res. 32 (1998) 19391947.
[2] O.W. Neiditch, J. Am, Oil Chem. Soc. 58 (1981) 162165.
[3] Ullmans Encyclopedia of Industrial Chemistry, 2001 Electronic release. 6th edition,
Wiley-VCH Interscience, 2001 (http://www.wiley-vch.de/contents/ullmann/ull_
10518.htm).
[4] Ciba Specialty Chemicals, Inc, Human and Environmental Risk Assessment on Ingredients of European Household Cleaning Products: Fluorescent Brightener FWA-1,
2004 (http://www.heraproject.com/RiskAssessment.cfm).
[5] Ciba Specialty Chemicals, Inc, Human and Environmental Risk Assessment on Ingredients of European Household Cleaning Products: Fluorescent Brightener FWA-5,
2004 (http://www.heraproject.com/RiskAssessment.cfm).
[6] C.J. Anastasiou, Optical Brightener Detection for Tracking Wastewater Contributions
to Coastal, Estuarine and Freshwater Systems, Department of Environmental Protection Watershed Management Program, Florida, 2007 130.
[7] Chinese National Standard (CNS) No. 12899, Identication of Fluorescent Brightening Agents Classes on Textiles, 1991 (http://www.webstore.jsa.or.jp/webstore/Com/
FlowControl.jsp).
[8] C.R. Ganz, C. Liebert, J. Schulze, S. Stensby, J. Water Pollut, Control Fed. 47 (1975)
28342849.
[9] A.W. Burg, C.J. Kensler, M.W. Rohovsky, Environ. Control 7 (1977) 91120.
[10] H. Hu, Z. He, Y. Ni, H. Zhang, Y. Zhou, Pulp Pap. 110 (2009) 3741.
[11] W. Coas, P.G. Desideri, L. Lepri, J. Chromatogr. 322 (1985) 363370.
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
117
J.L. Jasperse, P.H. Steiger, J. Am, Oil Chem. Soc. 69 (1992) 621625.
D. Kirkpatrick, J. Chromatogr. 139 (1977) 168173.
B.P. Mcpherson, N. Omelczenko, J. Am. Oil Chem. Soc. 57 (1980) 388391.
G. Calabro, P. Gurro, G. Micali, Analyst 109 (1984) 155158.
J.A. Field, T.M. Field, T. Poiger, W. Giger, Environ. Sci. Technol 30 (1996) 22202226.
W. Giger, J.M.A. Stoll, Anal. Chem. 69 (1997) 25942599.
W.H. Ding, W.C. Shu, J. Chromatogr. A 1088 (2005) 218223.
W.H. Ding, W.C. Shu, J. Chin. Chem. Soc. 56 (2009) 797803.
D.L. DuVal, K. Miyajima, I. Ogura, J. Am. Oil Chem. Soc. 72 (1995) 827833.
H.C. Chen, W.H. Ding, J. Chromatogr. A 1108 (2006) 202207.
L. Carolei, I.G.R. Gutz, Talanta 66 (2005) 118124.
S.H. Ahmadi, J. Azad, K. Kargosha, M. Mansourian, Talanta 75 (2008) 589593.
M.M. Reis, S.P. Gurden, A.K. Smilde, M.M.C. Ferreira, Anal. Chim. Acta 422 (2000)
2136.
R. Bro, P. Geladi, A.K. Smilde, J. Wiley, Multi-Way Analysis with Applications in the
Chemical Sciences, Wiley Online Library, Hoboken, NJ, 2004.
I.D. Wilson, U. Brikman, J. Chromatogr. 1000 (2003) 325335.
R. Bro, Anal. Chim. Acta 500 (2003) 185194.
Y. Akhlaghi, M.R. Hormozi-Nezhad, M. Kompany-Zareh, Anal. Chem. 84 (2012)
66036610.
D.M.F.Y. Ding, L.Q. Hu, H.L. Wu, A.L. Xia, R.Q. Yu, J. Chemom. 19 (2005) 6576.
L.Q. Hu, S.F. Li, H.L. Wu, A.L. Xia, R.Q. Yu, S.H. Zhu, J. Chemom. 21 (2007) 133144.
Z.P. Chen, J.H. Jiang, H.L. Wu, Y. Li, R. Yu, Q. Yu, Chemom. Intell. Lab. Syst. 52 (2000)
7586.
Z.P. Chen, H.L. Wu, R.Q. Yu, J. Chemom. 15 (2001) 439453.
http://www.models.kvl.dk/source.
A.C. Olivieri, H.L. Wu, R.Q. Yu, Chemom. Intell. Lab. Syst. 96 (2009) 246251.
A. Afkhami, M. Bahram, R. Bro, C. Stedmon, J. Chemometr. 20 (2006) 99105.
C.M. Andersen, A. Rinnan, Chemom. Intell. Lab. Syst. 76 (2005) 9199.
R. Bro, G. Tomasi, Chemom. Intell. Lab. Syst. 75 (2005) 163180.
R. Bro, K. Kjeldahl, H.A.L. Kiers, A.K. Smilde, Anal. Bioanal. Chem. 390 (2008)
12411251.
Practical Guide to the Analytical Validation of Body Fluid Chemistry Testing, Mayo
Medical Laboratories, 2013.
R.C.R. Beck, L.C. Cichota, R.B. Friedrich, A. Ravanello, C.M.B. Rolime, Quim Nova 321
(2009) 10521054.