Different approaches in Partial Least Squares and Artificial Neural Network models applied for the analysis of a ternary mixture of Amlodipine, Valsartan and Hydrochlorothiazide

Article abstract of DOI:10.1016/j.saa.2013.11.045

Different chemometric models were applied for the quantitative analysis of Amlodipine (AML), Valsartan (VAL) and Hydrochlorothiazide (HCT) in ternary mixture, namely, Partial Least Squares (PLS) as traditional chemometric model and Artificial Neural Networks (ANN) as advanced model. PLS and ANN were applied with and without variable selection procedure (Genetic Algorithm GA) and data compression procedure (Principal Component Analysis PCA). The chemometric methods applied are PLS-1, GA-PLS, ANN, GA-ANN and PCA-ANN. The methods were used for the quantitative analysis of the drugs in raw materials and pharmaceutical dosage form via handling the UV spectral data. A 3-factor 5-level experimental design was established resulting in 25 mixtures containing different ratios of the drugs. Fifteen mixtures were used as a calibration set and the other ten mixtures were used as validation set to validate the prediction ability of the suggested methods. The validity of the proposed methods was assessed using the standard addition technique.

Full text of DOI:10.1016/j.saa.2013.11.045

Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 122 (2014) 744–750
Contents lists available at ScienceDirect
Spectrochimica Acta Part A: Molecular and
Biomolecular Spectroscopy
journal homepage: www.elsevier.com/locate/saa
Different approaches in Partial Least Squares and Artificial Neural
Network models applied for the analysis of a ternary mixture of
Amlodipine, Valsartan and Hydrochlorothiazide
1
⇑
Hany W. Darwish , Said A. Hassan , Maissa Y. Salem, Badr A. El-Zeany
Department of Analytical Chemistry, Faculty of Pharmacy, Cairo University, Kasr El-Aini Street, 11562 Cairo, Egypt
h i g h l i g h t s
g r a p h i c a l a b s t r a c t
ꢀ
ꢀ
ꢀ
Advanced chemometric methods
developed for this ternary mixture.
Traditional (PLS) and advanced (ANN)
chemometric models.
Difference between GA and PCA as
preceding step to chemometric
models.
1) Raw data
2) GA model.
AML-1
AML-2
Genetic algorithm
ꢀ
GA can improve the prediction with
less LVs or neurons.
a r t i c l e i n f o
a b s t r a c t
Article history:
Received 10 May 2013
Received in revised form 10 September 2013
Accepted 3 November 2013
Available online 20 November 2013
Different chemometric models were applied for the quantitative analysis of Amlodipine (AML), Valsartan
VAL) and Hydrochlorothiazide (HCT) in ternary mixture, namely, Partial Least Squares (PLS) as tradi-
(
tional chemometric model and Artificial Neural Networks (ANN) as advanced model. PLS and ANN were
applied with and without variable selection procedure (Genetic Algorithm GA) and data compression
procedure (Principal Component Analysis PCA). The chemometric methods applied are PLS-1, GA-PLS,
ANN, GA-ANN and PCA-ANN. The methods were used for the quantitative analysis of the drugs in raw
materials and pharmaceutical dosage form via handling the UV spectral data. A 3-factor 5-level experi-
mental design was established resulting in 25 mixtures containing different ratios of the drugs. Fifteen
mixtures were used as a calibration set and the other ten mixtures were used as validation set to validate
the prediction ability of the suggested methods. The validity of the proposed methods was assessed using
the standard addition technique.
Keywords:
PLS
ANN
GA
Amlodipine
Valsartan
Hydrochlorothiazide
Ó 2013 Elsevier B.V. All rights reserved.
Introduction
AT
failure, and post-myocardial infarction [3].
Hydrochlorothiazide (HCT), 6-chloro-3,4-dihydro-2H-1,2,
1
-receptor. It is used for treatment of hypertension, heart
Amlodipine (AML), 2-[(2-aminoethoxy)methyl]-4-(2-chloro-
phenyl)-1,4-dihydro-6-methyl-3,5-pyridine
carboxylic
acid
4-benzothiadiazine-7-sulphonamide-1,1-dioxide [1] (Fig. 1c), is
a benzothiadiazines diuretic widely used in antihypertensive
pharmaceutical formulations [4].
Literature survey revealed that AML and HCT are official in
British Pharmacopoeia [5], while VAL, HCT and their mixture are
official in United States Pharmacopoeia [6]. There are many re-
ported methods for the determination of AML, VAL or HCT in differ-
ent dosage forms [7–14], but only few chromatographic methods
were reported for the simultaneous estimation of AML, VAL and
HCT in their ternary mixture [15–19]. Also, spectrophotometric
3
-ethyl 5-methyl ester) [1] (Fig. 1a) is a dihydropyridine derivative
acts as a calcium channel blocker. It is used in the management of
hypertension, stable angina and variant angina [2].
Valsartan (VAL), N-[p-(o-1H-Tetrazol-5-ylphenyl)benzyl]-N-
valeryl-L-valine [1] (Fig. 1b), is an antagonist of the angiotensin-II
⇑
Corresponding author. Tel.: +20 1000994542.
E-mail address: saidmonem_84@yahoo.com (S.A. Hassan).
Present address: Pharmaceutical Chemistry Department, College of Pharmacy,
1
King Saud University, P.O. Box 2457, Riyadh 11451, Saudi Arabia.
1
386-1425/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.saa.2013.11.045

H.W. Darwish et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 122 (2014) 744–750
745
Fig. 1. Structural formulae for (a) Amlodipine, (b) Valsartan, and (c) Hydrochlorothiazide.
and traditional chemometric methods were applied on this
mixture [20–22].
coefficient increase (Lci) control the variation of Lc value. It var-
ies as a function of performance of the ANN.
The rationales of this work were to:
Experimental
a. Develop simple and accurate methods for the simultaneous
determination of AML, VAL and HCT in their tablets.
b. Show the effect of variable selection (GA) and data compres-
sion (PCA) methods on enhancing the prediction power of
different chemometric models.
Materials and reagents
–
–
–
Amlodipine; kindly supplied by Al-Hekma pharmaceutical Com-
pany, Egypt, its purity was certified to be 99.9 ± 0.7.
Valsartan; kindly supplied by Novartis pharmaceutical Com-
pany, Egypt, its purity was certified to be 99.7 ± 0.2.
Hydrochlorothiazide; kindly supplied by Al-Hekma pharma-
ceutical Company, Egypt, its purity was certified to be
Neural networks
Artificial Neural Network (ANN) is a type of artificial intelli-
gence method that resembles biological nervous system in having
the ability to find the relationship between inputs and outputs. A
network is made up of a number of interconnected nodes (called
neurons) arranged into three basic layers (input, hidden and out-
put) that are interconnected by connections called weights. The
type of ANN used in this manuscript is feed-forward network
trained with the back propagation of errors learning algorithm.
The input nodes in this representation perform no computation
but are used to distribute inputs into the network. It is called
feed-forward ANN as information passes one way through the net-
work from the input layer, through the hidden layer and finally to
the output layer. The outputs (predicted concentrations) are com-
pared with targets (actual concentrations), and the difference be-
tween them is called error [23]. ANN parameters to be optimized:
9
9.8 ± 0.4.
EXFORGE HCT tablet dosage forms; labeled to contain 5(AML)/
60(VAL)/12.5(HCT) mg batch number 5002125, 5/160/25 mg
batch number 5002141 and 10/320/25 mg batch number
002159, manufactured by Novartis Pharmaceuticals Corpora-
Ò
–
1
5
tion, USA. They were procured from U.S.A. market.
–
Methanol; El-NASR Pharmaceutical Chemicals Co., Egypt.
Instruments
SHIMADZU dual beam UV–visible spectrophotometer (Kyoto/
Japan), model UV-1650 PC connected to IBM compatible and a
HP1020 laserjet printer. The bundled software, UV-Probe personal
spectroscopy software version 2.21 (SHIMADZU) was used. The
spectral band was 2 nm and scanning speed is 2800 nm/min with
–
The transfer functions: There are two transfer functions used in
ANN; one between input and output of a node in the hidden
layer and the other is applied in output layer. The use of these
functions depends on relationship between the inputs and out-
puts. Tansig-purelin transfer functions are commonly used for
non-linear systems [24] while purelin-purelin functions are
used for linear ones [25].
Hidden neurons number (HNN): It is related to the converging
performance of the output error function during the learning
process.
Number of neurons: Unfortunately, there are no fixed rules as to
how many neurons should be included in the hidden layer. If
there are too few nodes in the hidden layer the network may
have difficulty generalizing to problems it has never encoun-
tered before. On the other hand, if there are too many nodes
in the hidden layer, the network may take an unacceptably long
time to learn anything of any value.
Lc, Lcd and Lci: The learning coefficient (Lc) controls the degree
at which connection weights are modified during the learning
process. The learning coefficient decrease (Lcd) and learning
0
.1 nm interval.
Software
All chemometric methods were implemented in MatlabÒ
.0.0.19920 (R14). PLS, GA-PLS, ANN, GA-ANN and PCA-ANN were
7
carried out by using PLS toolbox software version 2.1 in conjunction
with Neural Network toolbox. The t-test and F-test were performed
using Microsoft Excel. All calculations were performed using a Dual
CPU, 1.47 GHz, 2.00 GB of RAM under Microsoft Windows Vista .
–
–
Ò
™
Procedures
Standard solutions
(
a) Standard stock solutions of AML, VAL and HCT 1 mg/mL in
methanol.
–
(
b) Standard working solutions for AML and VAL 80
for HCT 62.5 g/mL were prepared from stock solutions by
appropriate dilutions with methanol.
lg/mL and
l

7
46
H.W. Darwish et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 122 (2014) 744–750
Spectral characteristics of AML, VAL and HCT
The zero-order (D ) absorption spectra of AML, VAL and HCT
20 g/mL for each) solutions were recorded against methanol as
0
(
l
a blank over a range of 200–400 nm.
Experimental design for chemometric methods
A 5-level, 3-factor design was performed using 5 concentration
levels for each of the 3 compounds to be analyzed. The design
spans the mixture space fairly well; where there are 5 mixtures
for each compound at each concentration level, resulting in 25
mixtures [26]. The central level of the design is 6, 32 and
9
.375 lg/mL for AML, VAL and HCT, respectively. The concentra-
tion for each level for each compound is based on the calibration
range of each drug, the ratio of AML, VAL and HCT in the market
pharmaceutical product was involved. Table 1 represents the con-
centration design matrix. The regions from 200 to 230 nm were
rejected. Fifteen mixtures of this design were used as a calibration
set and the other 10 mixtures were used as a validation set to test
the predictive ability of the developed multivariate models.
Fig. 2. Zero order absorption spectrum of 20
20 lg/mL HCT (– – –) using methanol as blank.
lg/mL AML (. . .), 20 lg/mL VAL (–) and
tered into separate 100-mL measuring flasks, and the volume was
completed with methanol. Five mL aliquots were transferred into
Analysis of AML, VAL and HCT in EXFORGE HCTÒ tablets by the
proposed methods
1
0-mL measuring flasks and suitable aliquots of AML were
Five tablets of each Exforge HCTÒ formulation were accurately
weighed and finely powdered. An amount of the powder equiva-
lent to 8 mg VAL was weighed and dissolved in methanol by
shaking in ultrasonic bath for about 30 min. The solutions were fil-
transferred from their standard working solutions for spiking the
solutions to reach concentrations within linearity range and then
volumes were completed with methanol. The spectra of these
solutions were scanned from 200 to 400 nm, stored in the
computer and analyzed by the proposed methods.
Table 1
Results and discussion
The 5-level, 3-factor experimental design shown as concentrations of the mixture
components in
lg/mL.
The absorption spectra of the three compounds, AML, VAL and
HCT show highly overlapped spectral band in the region 200–
300 nm as shown in Fig. 2. Only AML can be determined by zero
order spectrophotometry.
Mix. No.
AML
6
10
6
10
2
4
2
8
4
10
6
8
4
10
6
8
10
8
2
8
2
VAL
32
28
24
32
24
36
40
40
40
36
28
32
28
40
40
24
24
36
36
28
32
36
24
28
32
HCT
9.375
15.625
3.125
6.25
15.625
15.625
6.25
1
.
a
2
.
The goal of this study was to develop accurate and simple che-
mometric methods for simultaneous determination of AML, VAL
and HCT in their pharmaceutical preparations and to present the
effect of data compression and variable selection procedures on
enhancing the predictive power of PLS and ANN models.
The first step in model building, involves constructing the cali-
bration matrix for the ternary mixture. In this study the model was
optimized with the aid of the 5-level 3-factor design [26] resulting
in 25 sample mixtures. Table 1 shows the composition of the 25
sample mixtures. These 25 sample mixtures were split to 15 train-
ing mixtures (for building the models) and 10 validation mixtures
(for measuring predictive power of the models).
The quality of multicomponent analysis is dependent on the
wavelength range and spectral mode used [27]. The wavelengths
used were in range 231–400 nm for AML, 231–320 nm for VAL
and 231–360 nm for HCT. Wavelengths less than 231 nm were re-
jected due to the noisy content. Wavelengths more than 320 nm
for VAL and 360 nm for HCT were not used because corresponding
compounds do not absorb in these regions.
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
.
.
.
.
.
.
.
0.
1.
2.
3.
4.
5.
6.
7.
8.
9.
0.
1.
2.
3.
4.
5.
12.5
9.375
9.375
6.25
15.625
12.5
3.125
15.625
9.375
12.5
6.25
3.125
3.125
12.5
12.5
6.25
Variable selection: Genetic Algorithm (GA)
Molecular spectroscopy has been greatly improved by the use of
variety of multivariate statistical methods [28,29]. Methods such
as Partial Least Squares (PLS) or principal component regression
6
4
2
4
(
PCR), allow to take into account the whole spectrum without
9.375
3.125
performing variable selection [30]. It has been recognized that an
efficient variable selection can be beneficial to improve the
a
The shaded rows represent the validation set

H.W. Darwish et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 122 (2014) 744–750
747
predictive ability of the model and to reduce its complexity [31].
Several techniques devoted to variable selection in PLS models ap-
plied to spectral data have been presented [32,33]. It has already
been shown that Genetic Algorithms (GAs) can be successfully
used as a variable selection technique [34,35]. The architecture of
a GA can be divided into five components: Initiation, Evaluation,
Exploitation, Exploration and Mutation. An important issue of suc-
cessful GA performance is the adjustment of GA parameters [36].
The fitness values were used as response variables. Mutation
rate was 0.005 in all cases as when it increased above this value,
no convergence occurred between average fitness and best fitness
values and model stop. The adjusted GA parameters are shown in
Table 2.
The GA was run on 170, 90 and 130 variables for AML, VAL and
HCT, respectively, using a PLS with maximum number of factors
allowed is the optimal number of components determined by
cross-validation on the model containing all the variables, and
the selected variables were used for running of PLS model and as
input data for ANN. GA reduced absorbance matrix to about
the optimum number of factors, which involves selecting that
model including the smallest number of factors that results in an
insignificant difference between the corresponding RMSECV and
the minimum RMSECV Fig. 3.
ANN
The large number of nodes in input layer of the network
(wavelength readings) increases the CPU time for ANN modeling,
the absorbance matrix was reduced either by Genetic Algorithm
(variable selection procedure) to about 36–46% of the original ma-
trix or Principal Component Analysis (PCA) (variable compression
procedure) to three principal components. Thus three ANNs
(ANN, GA-ANN and PC-ANN) were applied in our work. The output
layer is the concentration matrix of one component. The hidden
layer consists of just single layer which has been considered suffi-
cient to solve similar or more complex problems. Moreover, more
hidden layers may cause overfitting [25].
The values of the optimized ANN parameters for each drug are
shown in Table 3. From the most important parameters that should
be optimized carefully, transfer function pair. Choosing of transfer
function depends on the nature of data you work on. In our case,
purelin–purelin transfer function was implemented in our models
due to linear correlation between absorbance and concentration.
After optimization of architectures and parameters of the ANNs,
the training step was done. ANN was trained by different training
functions and there was no difference in performance (no decrease
in Mean Square Error MSE). Levenberg–Marquardt back propaga-
tion (TRAINLM) [42] was thus preferred as it is time saving. To
avoid overfitting of our model, the validation set was encountered
in training step and ANN stops when MSE of calibration set de-
creased and that of validation set increased. Analysis from raw
data, Genetic Algorithm model and Principal Component Analysis
was implemented to test for improvement of predictions.
The proposed chemometric methods were run on the calibra-
tion data using optimal parameters. The concentrations of the
three drugs in the calibration set (15 mixtures) were calculated.
By plotting predicted concentrations of each component versus
actual concentrations, a straight line was obtained (Table 1, Sup-
plementary Material). In order to validate the proposed methods,
the validation set (10 mixtures) was analyzed with the proposed
methods (Table 4).
3
6–46% of the original matrix (62, 42 and 54 wavelengths for
AML, VAL and HCT, respectively).
Partial Least Squares (PLS-1)
The purpose of PLS method is to build a calibration model be-
tween the concentration of the analytes under study (AML, VAL
and HCT in our case) and the latent variables of the data matrix
[
37]. Two different approaches can be used in Partial Least Squares
called PLS-1 and PLS-2. PLS-2 uses the whole information about the
concentration of all components to form latent variables (LVs),
while PLS-1 uses only the information about the concentration of
one component to create the LVs used by the model [38].
Including extra LVs in the model increases the possibility of the
known problem of overfitting. On the other hand, if the number of
LVs was too small meaningful data that could be necessary for the
calibration might be discarded. Therefore optimization of number
of the LVs is a critical issue in PLS method. Leave one out (LOO)
cross validation [39] and the bootstrap [40] can be applied to
predict the optimum number of PLS components.
PLS-1 method was run on the calibration data of absorption
spectra. To select the number of factors in the PLS-1 algorithm, a
cross validation (CV) method leaving out one sample at a time
The proposed PLS-1, GA-PLS, ANN, GA-ANN and PCA-ANN
[
41] was applied using calibration set of 15 calibration spectra.
methods were successfully applied for the determination of AML,
RMSECV (Root Mean Squares Error of Cross-Validation) indicates
both of the precision and accuracy of predictions. It was recalcu-
lated upon addition of each new factor to the PLS-1. The method
developed by Haaland and Thomas [28] was used for selecting
VAL and HCT in Exforge HCTÒ tablets, Table 5. The validity of the
proposed methods was further assessed by applying the standard
addition technique (Table 2, Supplementary Material).
The results obtained for the analysis of AML, VAL and HCT in Ex-
forge HCTÒ tablets by the suggested methods were statistically
compared with those obtained by applying the reported HPLC
method [16] and no significant difference between the results
was obtained as shown in Table 6.
Table 2
Parameters of the Genetic Algorithms.
GA reduced the optimal number of latent variables of PLS-1
model for AML from three into two factors. Also, recoveries and
RMSEP (Root Mean Square Error of Prediction) were decreased
indicating a better resolution power of GA-PLS model than PLS-1
model (Table 4).
GA allowed the use of less number of neurons (shorter training
time) for AML than those used in the network utilized the raw data.
While PCA-ANN did not show any improvement than ANN, even
the results were worse (Table 4). These results indicate that vari-
able selection models (GA) are more suitable than data compres-
sion procedure (PCA), when preceding ANN, for the analysis of
this ternary mixture. This result may be attributed to the fact
that GA introduces the most relevant wavelengths to the drug
concentration.
Parameter
Value
Population size
Maximum generations
Mutation rate
The number of variables in a window (window width)
Per cent of population the same at Convergence
20
50
0.005
2
100, except VAL
(
50)
%
Wavelengths used at initiation
50
Single
3, except AML (2)
Random
Crossover type
Maximum number of latent variables
Cross validation
Number of subsets to divide
Data into for cross validation
Number of iterations for cross validation at each
generation
4
2

7
48
H.W. Darwish et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 122 (2014) 744–750
AML-1
AML-2
VAL-1
VAL-2
HCT-1
HCT-2
Fig. 3. The optimum number of LV for AML, VAL and HCT concentration prediction from (1) raw data and from (2) GA model.
Table 3
Optimized parameters of ANNs.
Method
Drug
ANN
AML
GA-ANN
AML
PCA-ANN
AML
VAL
HCT
VAL
HCT
VAL
HCT
Architecture
Hidden neurons number
Transfer functions
Learning coefficient
Learning coefficient decrease
Learning coefficient increase
170-10-1
10
90-10-1
10
130-1-1
1
62-7-1
7
42-10-1
10
Purelin–Purelin
0.001
54-1-1
1
3-3-1
3
3-3-1
3
3-3-1
3
0.001
0.001
100
0.001
0.001
100
0.001
0.001
100
0.001
0.001
100
0.001
0.001
100
0.001
0.001
100
0.001
0.001
100
0.001
0.001
100
0.001
100

H.W. Darwish et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 122 (2014) 744–750
749
Table 4
Determination of AML, VAL and HCT in validation set by the proposed chemometric methods.
Concentration (
l
g/mL)
PLS-1
GA-PLS
ANN
GA-ANN
AML
PCA-ANN
AML
Recovery %a
AML VAL
HCT
AML
VAL
HCT
99.4
100.4
99.2
101.0
101.4
100.9
100.9
99.2
AML
VAL
99.9
HCT
99.3
101.5
99.4
100.9
100.4
101.0
101.0
100.0
100.0
100.4
AML
VAL
99.3
HCT
98.8
100.1
99.4
100.7
100.8
100.0
101.1
98.8
VAL
99.3
HCT
99.4
100.1
99.3
100.9
101.1
100.0
100.7
99.2
VAL
HCT
97.7
102.9
99.8
101.2
101.4
100.9
101.0
99.4
1
1
4
8
1
8
1
8
8
8
0
0
28
32
36
40
36
32
40
24
36
28
15.625
6.25
15.625
12.5
9.375
15.625
3.125
9.375 102.4
6.25
3.125
99.8
98.8
99.8
98.6
98.9
99.2
98.9
100.6
100.3
99.7
100.6
99.0
98.8
99.6
98.6
98.7
99.0
99.2
99.0
99.2
99.3
98.9
99.2
99.4
102.0
99.8
100.3
100.2
98.7
99.7
99.3
98.9
99.3
99.2
99.6
101.0
100.7
100.3
98.6
100.0
99.8
99.0
99.9
99.6
99.5
100.4
100.7
99.0
98.8
101.6
98.6
97.3
98.6
99.0
102.4
99.4
98.0
101.1
100.4
99.7
101.2
99.6
98.5
98.6
98.6
98.5
97.9
100.0
99.5
100.6
99.4
99.1
99.0
99.9
99.3
99.1
99.7
98.7
99.5
99.3
99.2
99.7
99.8
100.4
99.4
99.8
99.3
100.0
99.3
99.3
99.6
99.8
100.3
99.3
0
0
98.6
99.6
100.9
100.9
100.4
100.9
100.3
100.2
102.1
100.5
Mean
99.5
1.1
0.10
99.5
0.8
0.29
100.4
0.8
0.09
99.6
0.9
0.08
99.6
0.5
0.22
100.4
0.7
0.08
99.7
0.8
0.07
99.5
0.4
0.23
100.1
0.8
0.08
99.8
0.6
0.06
99.6
0.4
0.17
100.1
0.7
0.07
99.3
1.6
0.14
99.4
1.2
0.41
100.7
1.5
0.16
RSD%
RMSEPb
a
Average of three determinations.
Root Mean Square Error of Prediction.
b
Table 5
Determination of AML, VAL and HCT in Exforge HCT tablets by the proposed ANN and GA-ANN methods.
Ò
Product
Drug
PLS-1a
GA-PLSa
ANNa
GA-ANNa
PCA-ANNa
Exforge HCTÒ 5/160/12.5
AML
VAL
HCT
99.9 ± 0.8
100.1 ± 0.9
99.7 ± 0.9
100.1 ± 0.4
100.1 ± 0.6
100.0 ± 0.8
99.7 ± 1.0
100.1 ± 0.6
100.8 ± 0.6
99.9 ± 0.7
100.1 ± 0.7
100.2 ± 0.6
100.2 ± 1.1
99.6 ± 0.9
100.0 ± 0.6
Exforge HCTÒ 5/160/25
Exforge HCTÒ 5/160/25
AML
VAL
HCT
100.0 ± 0.8
100.0 ± 1.0
99.7 ± 0.8
99.6 ± 1.0
100.2 ± 0.8
99.6 ± 0.6
99.7 ± 1.2
100.5 ± 0.9
100.3 ± 1.0
99.8 ± 0.8
100.1 ± 0.8
99.8 ± 0.7
99.7 ± 0.8
100.0 ± 0.9
99.6 ± 1.0
AML
VAL
HCT
100.7 ± 1.1
99.5 ± 1.2
99.8 ± 0.9
100.7 ± 1.1
99.7 ± 1.1
99.6 ± 1.3
100.6 ± 1.1
99.7 ± 0.8
99.9 ± 0.8
100.3 ± 1.0
99.7 ± 1.0
99.9 ± 1.0
100.5 ± 1.1
99.7 ± 1.3
99.9 ± 1.8
a
Average of three determinations.
Table 6
Statistical comparison for the results obtained by the proposed methods and the reported method [16] for the analysis of AML, VAL and HCT in Exforge HCT tablets.
Ò
Value
PLS-1
Mean
RSD%
n
Variance
Student’s t testa (2.12)
F valuea (3.44)
AML
VAL
HCT
100.2
99.9
99.8
0.9
1.0
0.7
9
9
9
0.790
0.904
0.533
0.032
0.899
0.822
1.472
1.697
1.109
GA-PLS
AML
VAL
HCT
100.1
100.0
99.7
0.9
0.8
0.8
9
9
9
0.855
0.601
0.680
0.164
0.700
0.803
1.593
1.128
1.416
ANN
AML
VAL
HCT
100.0
100.1
100.3
1.0
0.7
0.8
9
9
9
1.031
0.558
0.657
0.457
0.401
0.866
1.920
1.047
1.369
GA-ANN
AML
VAL
HCT
100.0
100.0
99.9
0.8
0.8
0.7
9
9
9
0.594
0.586
0.511
0.618
0.693
0.257
1.107
1.099
1.064
PCA-ANN
Reported Methodb
AML
VAL
HCT
100.1
99.8
99.8
0.9
0.9
1.1
9
9
9
0.856
0.885
1.224
0.198
1.177
0.543
1.596
1.661
2.547
AML
VAL
HCT
100.2
100.2
100.0
0.7
0.7
0.7
9
9
9
0.537
0.533
0.480
–
–
a
The values in the parenthesis are the corresponding theoretical values of t and F at P = 0.05.
HPLC method using Luna C18 column, a mobile phase consisting of methanol – phosphate buffer (30 mM, pH 5.5) (62:38 by volume) at a flow rate of 1 mL/min and UV
b
detection at 234 nm.
ANN gave better recoveries & RMSEP than PLS-1 (Table 4),
which may be due to the fact that ANN is a type of artificial
intelligence and that in ANN there is no chance for overfitting that
may occur in PLS calibrations.
The proposed chemometric methods show higher sensitivity
over the sequential spectrophotometric method [22] and better
mean recovery and RMSEP than the traditional methods applied
in our previous work [20] being advanced methods using artificial

7
50
H.W. Darwish et al. / Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 122 (2014) 744–750
intelligence and preceded by variable selection or data compres-
sion procedures.
[13] N. Devanaboyina, T. Satyanarayana, B.G. Rao, Int. J. Pharma Bio Sci. 3 (2012)
07–115.
14] P. Antil, D. Kaushik, G. Jain, K. Srinivas, I. Thakur, J. Chromatogr. Sep. Tech.
2013), http://dx.doi.org/10.4172/2157-7064.1000182.
15] H.W. Darwish, S.A. Hassan, M.Y. Salem, B.A. El-Zeany, Int. J. Pharma Bio Sci. 4
2013) 345–356.
1
[
[
[
(
Conclusion
(
16] S.E. Vignaduzzo, P.M. Castellano, T.S. Kaufman, J. Liq. Chromatogr. Relat.
Technol. 34 (2011) 2383–2395.
Five chemometric methods (PLS-1, GA-PLS, ANN, GA-ANN and
PCA-ANN) have been presented as powerful chemometric methods
to resolve the ternary mixture of AML, VAL and HCT in their
powder and pharmaceutical dosage forms. The effect of GA and
PCA as preceding step for chemometrics was studied. The results
in this paper suggested that the proposed methods can be
classified among accurate and sensitive methods. These merits
show the possibility to use the proposed methods in quality
control analysis of AML, VAL and HCT in laboratories lacking liquid
chromatographic instruments.
[
[
17] R.N. Sharma, S.S. Pancholi, Acta Pharm. 62 (2012) 45–58.
18] S.M. El-Gizawy, O.H. Abdelmageed, M.A. Omar, S.M. Deryea, A.M. Abdel-
Megied, Am. J. Anal. Chem. 3 (2012) 422–430.
[19] K. Anandakumar, D. Jothieswari, R. Subathrai, D. Santhi, T. Vetrichelvan, Acta
Chromatogr. 24 (2012) 37–50.
[
[
[
20] H.W. Darwish, S.A. Hassan, M.Y. Salem, B.A. El-Zeany, Spectrochim. Acta Part A
13 (2013) 215–223.
1
21] K. Ananda, M. Jayamariappan, Int. J. Pharm. Pharm. Sci.
27.
3 (2011) 23–
22] H.W. Darwish, S.A. Hassan, M.Y. Salem, B.A. El-Zeany, Int. J. Spectrosc., 2013,
in press), http://dx.doi.org/10.1155/2013/273102.
(
[
[
23] C.W. Dawson, R. Wilby, Hydrol. Sci. J. 43 (1998) 47–66.
24] A. Abbaspour, L. Baramakeh, Spectrochim. Acta Part
A
64 (2006) 477–
4
82.
Appendix A. Supplementary material
[
25] A. Khanchi, M. Mahani, M. Hajihosseini, M. Maragheh, M. Chaloosi, F. Bani,
Food Chem. 103 (2007) 1062–1068.
[
[
26] R.G. Brereton, Analyst 122 (1997) 1521–1529.
27] M. Blanco, J. Coello, F. Gonzalez, H. Iturriaga, S. Maspoch, J. Pharm. Sci. 82
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.saa.2013.11.045.
(
1993) 834–837.
[
28] D.M. Haaland, E.V. Thomas, Anal. Chem. 60 (1988) 1193–1202.
References
[29] W. Lindberg, J. Persson, S. Wold, Anal. Chem. 55 (1983) 643–648.
[
[
[
30] E.V. Thomas, D.M. Haaland, Anal. Chem. 62 (1990) 1091–1099.
31] E.V. Thomas, Anal. Chem. 66 (1994) 795–804.
32] F. Lindgren, P. Geladi, S. Rännar, S. Wold, J. Chemom. 8 (2005) 349–363.
[
[
1] A.C. Moffat, M.D. Osselton, B. Widdop, Clarke’s Analysis of Drugs and Poisons,
Pharmaceutical Press, London, 2004.
2] G.K. McEvoy, American Hospital Formulary Service , American Society of
Ò
[33] M. Forina, C. Casolino, C. Pizarro Millan, J. Chemom. 13 (1999) 165–
84.
1
Health-System Pharmacists Inc., Bethesda, 2001.
[
34] L. Davis, M. Mitchell, Handbook of Genetic Algorithms, Van Nostrand Reinhold,
New York, 1991.
[
[
3] R. Dina, M. Jafari, Am. J. Health-Syst. Pharm. 57 (2000) 1231–1241.
4] L.J. Brunton, K.L. Parker, Goodman & Gilman’s The Pharmacological Basis of
Therapeutics, 11 ed., McGraw-Hill, New York, 2006.
[
35] Z. Michalewicz, Genetic Algorithms+Data Structures, 3 ed., Springer, Berlin,
1996.
[
[
[
5] British Pharmacopoeia, Medicines and Healthcare Products Regulatory Agency
[
[
36] T. Li, C. Lucasius, G. Kateman, Anal. Chim. Acta. 268 (1992) 123–134.
37] R. Kramer, Chemometric Techniques for Quantitative Analysis, Marcel Dekker
Inc., New York, 1998.
(
MHRA), London, 2007.
6] United States Pharmacopeia, United States Pharmacopeial Convention, Rockville,
007.
7] H.W. Darwish, S.A. Hassan, M.Y. Salem, B.A. El-Zeany, Spectrochim. Acta Part A
04 (2013) 70–76.
2
[
38] D. Massart, B. Vandeginste, L. Buydens, S. De Jong, P. Lewi, J. Smeyers-Verbeke,
C.K. Mann, Handbook of Chemometrics and Qualimetrics: Part A, Elsevier,
Amsterdam, 1998.
1
[
[
8] M.M. Parambi, V. Ganesan, J. Appl. Pharm. Sci. 1 (2011) 97–99.
9] N.K. Ramadan, H.M. Mohamed, A.A. Mostafa, Port. Electrochim. Acta 30 (2012)
[
[
39] S. Wold, Technometrics (1978) 397–405.
40] B. Efron, R. Tibshirani, An Introduction to the Bootstrap, Chapman & Hall/CRC,
1
5–29.
10] H.W. Darwish, S.A. Hassan, M.Y. Salem, B.A. El-Zeiny, Spectrochim. Acta Part A
3 (2011) 140–148.
1
993.
41] A. Espinosa-Mansilla, F. Salinas, I. De Orbe Paya, Anal. Chim. Acta. 313 (1995)
03–112.
42] J. Torrecilla, M. Mena, P. Yáñez-Sedeño, J. García, J. Food Eng. 81 (2007) 544–
52.
[
[
[
8
1
[
[
11] V. Vichare, V. Tambe, V. Kashikar, D. SN, Int. J. Chem. 2 (2011) 7–10.
12] H.W. Darwish, S.A. Hassan, M.Y. Salem, B.A. El-Zeany, Int. J. Pharma Bio Sci. 4
5
(
2013) 230–243.