Computer-aided design of novel antibacterial 3-hydroxypyridine-4-ones: Application of QSAR methods based on the MOLMAP approach

Article abstract of DOI:10.1007/s10822-012-9561-2

3-Hydroxypyridine-4-one derivatives have shown good inhibitory activity against bacterial strains. In this work we report the application of MOLMAP descriptors based on empirical physicochemical properties with genetic algorithm partial least squares (GA-PLS) and counter propagation artificial neural networks (CP-ANN) methods to propose some novel 3-hydroxypyridine-4-one derivatives with improved antibacterial activity against Staphylococcus aureus. A large collection of 302 novel derivatives of this chemical scaffold was selected for this purpose. The activity classes of these compounds were determined using the two quantitative structure activity relationships models. To evaluate the predictability and accuracy of the obtained models, nineteen compounds belonging to all three activity classes were prepared and the activity of them was determined against S. aureus. Comparing the experimental results and the predicted activity classes revealed the accuracy of the obtained models. Seventeen of the nineteen synthesized molecules were correctly predicted by GA-PLS model according to the antimicrobial evaluation method. Molecules 5f and 5h proved to be moderately active and active experimentally, but were predicted as inactive and moderately active compounds, respectively by this model. The CP-ANN based prediction was correct for sixteen out of the nineteen synthesized molecules. 5a, 5h and 5q were moderately active and active based on the antimicrobial assays, but they were introduced as members of inactive, moderately active and inactive classes of compounds, respectively according to CP-ANN model. Springer Science+Business Media B.V. 2012.

Full text of DOI:10.1007/s10822-012-9561-2

J Comput Aided Mol Des (2012) 26:349–361
DOI 10.1007/s10822-012-9561-2
Computer-aided design of novel antibacterial 3-hydroxypyridine-
4-ones: application of QSAR methods based on the MOLMAP
approach
•
Razieh Sabet Afshin Fassihi Bahram Hemmateenejad
•
•
•
Lotfollah Saghaei Ramin Miri
•
Maryam Gholami
Received: 10 December 2011 / Accepted: 8 March 2012 / Published online: 28 March 2012
Ó Springer Science+Business Media B.V. 2012
Abstract 3-Hydroxypyridine-4-one derivatives have
shown good inhibitory activity against bacterial strains. In
this work we report the application of MOLMAP
descriptors based on empirical physicochemical properties
with genetic algorithm partial least squares (GA-PLS) and
counter propagation artificial neural networks (CP-ANN)
methods to propose some novel 3-hydroxypyridine-4-one
derivatives with improved antibacterial activity against
Staphylococcus aureus. A large collection of 302 novel
derivatives of this chemical scaffold was selected for this
purpose. The activity classes of these compounds were
determined using the two quantitative structure activity
relationships models. To evaluate the predictability and
accuracy of the obtained models, nineteen compounds
belonging to all three activity classes were prepared and the
activity of them was determined against S. aureus. Com-
paring the experimental results and the predicted activity
classes revealed the accuracy of the obtained models.
Seventeen of the nineteen synthesized molecules were
correctly predicted by GA-PLS model according to the
antimicrobial evaluation method. Molecules 5f and 5h
proved to be moderately active and active experimentally,
but were predicted as inactive and moderately active
compounds, respectively by this model. The CP-ANN
based prediction was correct for sixteen out of the nineteen
synthesized molecules. 5a, 5h and 5q were moderately
active and active based on the antimicrobial assays, but
they were introduced as members of inactive, moderately
active and inactive classes of compounds, respectively
according to CP-ANN model.
Keywords Antibacterial activity ꢀ CP-ANN
3-Hydroxypyridin-4-one ꢀ MOLMAP ꢀ QSAR
Abbreviations
ANN
Artificial neural networks
Counter propagation artificial neural
networks
CP-ANN
2D
Two dimensional
3D
Three dimensional
Electronic supplementary material The online version of this
article (doi:10.1007/s10822-012-9561-2) contains supplementary
material, which is available to authorized users.
GA-PLS
MIC
MLR
Genetic algorithm partial least squares
Minimum inhibitory concentration
Multiple linear regression
R. Sabet ꢀ A. Fassihi (&) ꢀ L. Saghaei
MOLMAP Molecular map of atom-level properties
Department of Medicinal Chemistry, School of Pharmacy and
Pharmaceutical Sciences and Isfahan Pharmaceutical Research
Center, Isfahan University of Medical Sciences, Isfahan, Iran
e-mail: fassihi@pharm.mui.ac.ir
SOM
Self-organizing maps
QSAR
Quantitative structure activity relationships
R. Sabet
Department of Medicinal Chemistry, Faculty of Pharmacy,
Shiraz University of Medical Sciences, 7146864685 Shiraz, Iran
Introduction
Iron is one of the trace elements inevitably required for the
survival and proliferation of all living things [1]. Pathogenic
bacteria require iron to proliferate and cause infectious dis-
eases in the human body. Vertebrate hosts withhold iron
B. Hemmateenejad (&) ꢀ R. Miri ꢀ M. Gholami
Medicinal & Natural Product Chemistry Research Center, Shiraz
University of Medical Sciences, Shiraz, Iran
e-mail: hemmatb@sums.ac.ir
123

350
J Comput Aided Mol Des (2012) 26:349–361
from microbial invaders as a major defence mechanism
against infection [2]. Many bacteria synthesize small mole-
cules known as siderophores, which possess high affinity to
iron and scavenge it from the environment [3]. 3-Hydroxy-
pyridine-4-one derivatives and their oxygen-containing
analogues, 3-hydroxypyran-4-ones, belong to a class of iron
chelators with reported in vitro antibacterial, antifungal and
antimalarial activities [4–7]. They have shown inhibitory
effects on the growth of Escherichia coli, Listeria inocua and
Staphylococcus aureus [8–10]. We have previously reported
the antimicrobial activities of novel Mannich bases of
2-alkyl-3-hydroxy-pyridine-4-ones as well as 1-alkyl
substituted 2-alkyl-3-hydroxypyridine-4-ones. They have
shown inhibitory effects on the growth of S. aureus, Sal-
monella enteritidis and Aspergillus flavus [11].
activated neurons is a map of the reactivity features of that
molecule (MOLMAP) as a fingerprint of the bonds avail-
able in that structure.
Regression models for a series of 3-hydroxypyrane-4-
one and 3-hydroxypyridine-4-one derivatives have been
previously investigated by our research group [11, 22].
According to these studies the most significant QSAR
models were obtained by GA-PLS against S. aureus.
Sometimes biological data can be classified into discrete
categories. For example, a chemical may be classified as
either active or inactive, or in several classes according to
the potency of the activity. In such cases, other statistical
techniques, such as classification methods must be applied,
in which the physicochemical properties of the compounds
are used to discriminate between activity and inactivity.
In this paper, we used the PETRA 3.20 software pack-
age to calculate empirical physicochemical properties of
bonds, and MOLMAP descriptors were generated on the
basis of the bond properties [19, 20]. These descriptors
were applied with GA-PLS and CP-ANN methods to pro-
pose some novel 3-hydroxypyridine-4-one derivatives with
improved antibacterial activity against S. aureus. The
proposed compounds were prepared in the lab and their
antibacterial potency was evaluated. The reason for
selecting this microorganism and this chemical scaffold
was the good diversity in the antimicrobial potencies
observed in our previously reported studies. Significant
GA-PLS models have described the effect of structural
modification of this scaffold and anti S. aureus activity too
[11, 22]. We performed the computational part of the study
on a large collection consisted of 302 novel derivatives of
3-hydroxypyridine-4-one. Most of these compounds have
not been prepared experimentally yet and none were
evaluated as antimicrobial agents previously. It is well
known that the hydrazone group plays an important role for
the antimicrobial activity in many different scaffolds.
Furthermore, a number of acylhydrazone derivatives have
been demonstrated to possess interesting antibacterial,
antifungal [23], antimalarial [24] and antitubercular activ-
ities [25]. Schiff’s bases have also been shown to exhibit a
broad range of biological activities, including antiviral,
antimalarial, antifungal and antibacterial properties [26,
27]. Thus, the antimicrobial capacity of hydrazide-hydra-
zone and Schiff’s base moieties was considered in the
design of these compounds.
Quantitative structure activity relationships (QSAR)
approach has been widely developed because of its powerful
ability to predict drug activity [12]. QSAR models are math-
ematical equations relating chemical structures to their bio-
logical activities. Various methods have been applied to
construct QSAR models including linear and nonlinear
regression methods. Multiple linear regression (MLR) and
artificial neural networks (ANN) have been extensively
employed in QSAR studies owing to their outstanding linear
and nonlinear mapping capability, respectively [13, 14].
Appropriate application of the structural and physicochemical
features of molecules is an essential key to achieve successful
QSAR models [12]. Nowadays, a multi-way analytical
method based on Kohonen network, originated from a method
for calculation of molecular descriptors called MOLMAP
(molecularmapofatom-levelproperties), hasbeenintroduced
[15–17]. Kohonen self-organizing maps (SOM) are a class of
unsupervised neural networks whose characteristic feature is
the reduction of multidimensional data to 2 dimensional (2D)
ones [18]. In QSAR modeling based on MOLMAP approach,
the resulted Kohonen scores are used as descriptors for clas-
sification and regression purposes.
The chemical reactivity of a compound, being related to
the ability to make and break bonds, primarily depends on
the properties of the bonds available in a molecule. Ga-
steiger et al. proposed that seven empirical physicochem-
ical properties are particularly relevant for representing
bonds and for modeling chemical reactivity [19]. These
properties are calculated by empirical procedures imple-
mented in PETRA program [20]. To explore all that
information for an entire molecule, and simultaneously
have a fixed-length representation, they proposed to map
all the bonds of a molecule into a fixed-length 2D SOM: a
MOLMAP [21]. A SOM must be trained previously with a
diversity of bonds from different structures (each bond
described by the seven bond properties) [18]. Then, all the
bonds of one molecule are submitted to the trained SOM,
each bond activates one neuron, and the pattern of
Materials and methods
QSAR analysis
The experiments described here required two major steps:
generation of descriptors and development of predictive
123

J Comput Aided Mol Des (2012) 26:349–361
351
models. To generate the descriptors, some properties were
processed by a Kohonen SOM to MOLMAPs scores [18].
For training this map, each object of the training set was a
chemical bond, each bond represented by seven empirical
physicochemical properties. This training set contains
bonds from diverse structures. Once trained, the map is
used to obtain molecular descriptors. Bonds of one mole-
cule were submitted to the trained SOM, and the pattern of
activated neurons, which is a map of the reactivity features
of that molecule (MOLMAP), is the descriptor of the
molecule. The second step consisted of establishing rela-
tionships between MOLMAP descriptors and antimicrobial
activity against S. aureus.
bond, and its weights were then adjusted to make them
even more similar to the properties of the presented bond.
Not only the winning neuron, but also the neurons in its
neighborhood have their weights adjusted. The extent of
adjustment depends, however, on the topological distance
to the winning neuron, the closer a neuron is to the central
neuron the larger is the adjustment of its weights. The
objects of the training set are iteratively fed to the map, the
weights corrected, and the training is stopped when a pre-
defined number of cycles are attained. A trained Kohonen
SOM will reveal similarities in the objects of a data set in
the sense that similar objects (similar bonds) are mapped
into the same or closely adjacent neurons [18]. Figure 1s is
shown in the supplementary materials.
Data set
Molecular descriptors
A set of thirty-one 3-hydroxypyridine-4-one and 3-hy-
droxypyran-4-one derivatives was compiled from literature
and our previous reports [4, 5, 11]. The antimicrobial
activity (against S. aureus) was reported as MIC. Different
strains of this microorganism were used in these reports:
PTCC 1337, ATCC 25923, and PTCC 29213. It was
assumed that there will be no big differences between the
susceptibility of these different strains and the strain
applied in the present study (PTCC 1112) against the
antimicrobial activity of the studied compounds. The
molecules were classified as active (class 1) if they
exhibited 8 lg/mL B MIC B 32 lg/mL (3 molecules);
moderately active (class 2) if 64 lg/mL B MIC B 128 lg/
mL (9 molecules) and inactive (class 3) if MIC was more
than 128 lg/mL (19 molecules). The structural features,
biological activity and the class of these compounds are
listed in Table 1.
Chemical bonds were represented by seven empirical
physicochemical bond properties calculated by PETRA
3.20. These properties were the difference of r electro-
negativity between the two atoms of the bonds, difference
of p atomic charges, difference of total atomic charges,
bond polarity, mean bond polarizability, resonance stabil-
ization, and bond dissociation energy. As some properties
depend on the orientation of the bond, each bond was
represented as (A–B) [21]. Figure 1 shows the common
bonds (11 bonds) in the molecules and the numbers
assigned to the bonds. A three-dimensional array of
empirical physicochemical parameters (molecules, bonds
and the empirical physicochemical properties on each
dimension, a 333 9 11 9 7 array) was provided (Fig. 1).
Then, the pattern of the activated neurons can be consid-
ered as a fingerprint of the objects and constitute their
MOLMAP scores [18]. At last; the map is transformed into
a vector by concentration of columns resulting in a fixed
length MOLMAP score where each scores of each object
have a dimension of (N 9 N). The best value of v was
obtained by trial and error, and the best results were
obtained for N = 5. It should be noted that output layer
dimensions (4 9 4) to (13 9 13) over (10–120) epochs
was also checked but the best results were achieved using
5 9 5 = 25 over 20 epochs. For Kohonen mapping, the
MOLMAP toolbox, developed by Milano Chemometrics
and QSAR research Group, was used [28].
Training of a Kohonen self-organizing map with empirical
physicochemical descriptors
Kohonen SOM is an unsupervised neural network that
reveals similarities between objects. It can be used for the
reduction of multidimensional objects to two dimensional
objects. In this study, we used SOMs to reduce the
dimensions of chemical bonds, represented by seven
empirical physicochemical bond properties calculated by
PETRA to 2D data. A Kohonen SOM consists of a grid of
so-called neurons, each containing as many elements
(weights) as input variables. Figure 1s shows the archi-
tecture of a Kohonen network. Here the objects are bonds,
and the input variables are the seven properties of bonds.
Before the training starts, the weights take random values.
During the training, each individual bond is mapped into
the neuron that contains the most similar weights compared
to its properties. This is the central neuron, or winning
neuron [18]. The winning neuron was activated by the
Modeling procedures
Model development with PLS In this approach the partial
least squares (PLS) regression was employed to evaluate
the structure–activity relationships and genetic algorithm
(GA) was used as variable selection method. In the case of
MOLMAP approach the set of 25 Kohonen scores (i.e.,
5 9 5 Nodes) over 60 epochs was used as input. In order to
123

352
J Comput Aided Mol Des (2012) 26:349–361
Table 1 Chemical structure, experimental and predicted class of the antimicrobial activity of compounds used in QSAR analysis against S.
aureus by GA-PLS and CP-ANN
O
R3
R2
R1
R4
X
Compd.
X
R1
R2
R3
R4
Experimental
MIC (lg/mL)
Activity class
of compound
GA-PLS
CP-ANN
1
NH
CH3
OH
CH2–Ra
CH2–Ra
H
64
512
512
512
512
512
128
512
512
128
256
256
256
256
64
2
3
3
3
3
3
2
3
3
2
3
3
3
3
2
3
3
3
2
2
3
1
3
3
2
2
1
2
3
1
3
2
3
3
3
3
3
2
3
3
2
3
3
3
3
2
3
3
3
3
1
3
1
3
3
2
2
1
2
3
1
3
2
3
3
3
3
3
2
3
3
2
3
3
3
3
2
3
3
3
3
3
3
3
3
3
2
2
1
2
1
1
3
2
NH
C2H5
OH
H
3
NH
CH3
OH
CH2–N(CH3)2
H
4
NH
C2H5
OH
CH2–N(CH3)2
H
5*
6
NH
CH3
OH
CH2–N(C2H5)2
H
NH
C2H5
OH
CH2–N(C2H5)2
H
7
N–Ph
CH3
OH
H
H
8*
9
N–m–OH–Ph
CH3
OH
H
H
N–C3H7
CH3
OH
H
H
10
11
12
13
14
15
16
17
18
19*
20
21
22
23
24*
25
26
27
28*
29
30*
31
a
N–C4H9
CH3
OH
H
H
O
CH2Cl
CH3
H
OH
H
O
H
OH
H
O
CH2OH
CH2OH
CHO
OH
H
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
H
O
OCH2Ph
OCH2Ph
OCH2Ph
OCH2Ph
OCH2Ph
OCH2Ph
OH
H
O
H
O
COOH
CONHRb
CONHRc
CONHRd
CONHRb
CONHRc
CONHRd
CH2OH
COOH
CONHPh
CONHPh
CONHPh
CONH–Re
CONH–Re
CONH–Re
CH2OH
H
256
256
256
128
64
O
H
O
H
O
H
O
H
O
OH
H
256
8
O
OH
H
O
H
OCH2Ph
OCH2Ph
OCH2Ph
OCH2Ph
OH
256
256
128
128
16
O
H
H
O
H
H
N–CH3
N–CH3
O
H
H
H
H
H
OCH2Ph
OCH2Ph
OH
H
128
256
32
N–CH3
N–CH3
O
H
H
H
H
H
OH
H
256
Compounds used as prediction set
CH3
N
, Rb is
Ra is
, Rc is
, Rd is
, Re is
N
N
O
O
investigate the prediction ability of the models, the data set
(n = 31) was divided into two group: calibration set
(n = 25) and prediction set (n = 6). Given 25 calibration
samples; leave-one out cross-validation procedure was
used to find the optimum number of latent variables for
each PLS model. GA produces a population of acceptable
123

J Comput Aided Mol Des (2012) 26:349–361
353
Fig. 1 Common bonds and the
number of bonds in the
molecules
O
X
Electron-impact ionization was performed at an ionizing
energy of 70 eV. The synthesis pathway followed for the
preparation of the compounds is represented in Scheme 1.
The commercially available maltol was chosen as starting
compound to design several novel acylhydrazone deriva-
tives. 3-(3-hydroxy-2-methyl-4-oxopyridin-1(4H)-yl) ben-
zoic acid (2) was prepared by the reaction of maltol (1) and
3-aminobenzoic acid in aqueous ethanol at pH 5.0 [31].
Compounds 3 and 4 were prepared according to our recent
report [32] as follows. Carboxylic acid 2 was treated in
methanol with carbonyl diimidazole (DCI) and dimethyl
aminopyridine (DMAP) to furnish the desired methyl 3-(3-
hydroxy-2-methyl-4-oxopyridin-1(4H)-yl) benzoate (3)
derivative. Refluxing 3 with hydrazine-hydrate in methanol
produced 3-(3-hydroxy-2-methyl-4-oxopyridin-1(4H)-yl)
benzohydrazide (4), which was condensed with an aro-
matic aldehyde to yield (5a–o). Compounds 5p–s were
prepared by the same method and recently reported by this
group [32].
5
R3
R4
OH
R1
6
10
7
3
4
2
11
1
9
8
models in each run. In this work, many different GA-PLS
runs were conducted using different initial set of popula-
tions (50–250) and therefore a large number of acceptable
models were created.
The PLS regression method used was the NIPALS-
based algorithm existed in the chemometrics toolbox of
MATLAB software (version 7.1 Math work Inc.). Leave-
one-out cross-validation procedure was used to obtain the
optimum number of factors based on the Haaland and
Thomas F-ratio criterion [29].
Model development with CP-ANNs Genetic algorithm
(GA) was used for variable selection. Counter-propagation
artificial neural networks (CP-ANNs) were used to model
the relationship between the MOLMAP descriptors of the
compounds and the corresponding MIC values. Four
MOLMAP descriptors selected by genetic algorithm were
used to train CP-ANNs of size 7 9 7 nodes over 60
epochs. Leave-one-out (LOO) method was used to predict
the class of the compounds. CP-ANNs are very similar to
Kohonen maps and are essentially based on the Kohonen
approach, but combine characteristics from both supervised
and unsupervised learning [30]. CP-ANNs consist of two
layers of neurons. The first layer is a Kohonen map, and
this is responsible for choosing the winning neuron. It
stores information concerning the input data. The second
layer; whose neurons have as many weights as the number
of classes to be modeled, when dealing with classification
issues, have exactly the same number and the same layout
of neurons as the Kohonen one. CP-ANN generally is a
successful method for modeling classes separated with
non-linear boundaries [30].
General procedure for the synthesis of 3-[3-Hydroxy-2-
methyl-4-oxopyridine-1(4H)-yl] benzoic acid [aryl (het-
aryl) methylene]-hydrazides (5a–o).
Equimolar amounts of 4 and appropriate aldehyde
(1.00 mmol) in 20 mL ethanol were heated under reflux for
24 h. The obtained precipitate was filtered and recrystal-
lized twice from ethanol/diethyl ether (compounds 5a, 5e,
5g, 5m, 5p, 5q, 5r, and 5s). The other compounds (5b, 5c,
5d, 5f, 5h, 5i, 5j, 5k, 5l, 5n, and 5o) were purified by
preparative thin layer chromatography using (chloroform/
methanol:100/8).
Antimicrobial activity determination
Mueller–Hinton broth at pH 7.0 was used to culture S.
aureus [33]. The inoculum of microorganism (108 c.f.u/
mL) was cultured for 16–24 h at 37 °C and prepared to
turbidity equivalent to McFarland standard No. 0.5 The
final size of inoculum was 1.5 9 104. The test compounds
were dissolved in dimethyl sulfoxide (DMSO) (final con-
centration of 0.5 % v/v) and diluted with culture broth to
concentration of 512 lg/mL. Serial two fold dilutions were
made in concentration range from 1 to 512 lg/mL in sterile
96-well microplates. 100 lL of each dilution were dis-
tributed in 96-well microplates, as well as a sterility control
and a growth control (containing culture broth plus DMSO,
without antimicrobial substance). 100 lL of a 24 h old
inoculum of 105 c.f.u/mL was added to each well. Plates
were covered and sealed with parafilm and incubated for
24 h at 37 °C. MIC values were defined as the lowest
concentration of each tested compound, which completely
inhibited microbial growth [34]. Ampicilin and tetracycline
were used as standard antibacterial drugs.
Chemistry
All chemicals used for the synthesis of the compounds
were purchase from Merck or Fluka. Melting points were
determined on a Mettler capillary melting point apparatus
and were uncorrected. The IR spectra were recorded on a
WQF-510 Ratio Recording FT-IR spectrometer as a KBr
disc (c, cm-1). The 1H-NMR spectra (DMSO-d6) were
recorded on a Bruker 400 MHz spectrometer. Chemical
shifts (d) are reported in ppm downfield from the internal
standard tetramethylsilane (TMS). The mass spectra were
acquired with a Finnigan TSQ-70 mass spectrometer.
123

354
J Comput Aided Mol Des (2012) 26:349–361
NH2
COOH
O
N
O
N
O
O
O
OH
OH
OH ArCHO
OH
CH3OH
NH2NH2,H2O
C2H5OH
CDI,DMAP
C2H5OH
CH3
CH3
CH3
N
CH3
pH5
COOH
COOCH3
CONHNH2
1
2
3
4
O
1
6 OH
2
3
4
7
CH3
5
N
8
13
9
Cl
3'Br
2'
2' 3'
2'
2'
2'
2'
3'
H
N
15
17
3'
3'
3'
14
10
4'
4'
Ar:
Ar
4'Br
4'Cl
4'NO2
4'CH3
12
1'
1'
1'
1'
1'
1'
11
16
6' 5'
O
6' 5'
6' 5'
6' 5'
6' 5'
6' 5'
5a-s
5c
5e
5a
5b
5d
5f
H3CO
3'
N
3'
N
2'
4' 5'
4'
1'
4'
3'
2' 3'
3' 4'
3'
2'
2'
2'
4'
6'
7'
4'OCH3
1'
5' NO2
5' SCH3
2'
1'
5' NO2
N
N
O
8'
6' 5'
1'CH3
1C' H3
6' 5'
1'
5i
5g
5h
5j
5l
5k
2' 3'NO2
4'
HO
4'
2'
4'
3'
2'
3'
2'
2' 3'
3'
3'
1'
4' -CH3
4'
5'
5'
4'N(CH3)2
1'
1'
1'
1'
2'
O
S
6' 5'
6' 5'
6' 5'
1'
6' 5'
1'
5m
5q
5n
5o
5p
5r
5s
Scheme 1 Synthesis of N-aryl-3-hydroxypyridine-4-ones 5a–s
Results and discussions
first row are numbered as N1–N5, those of the second row
as N6–N12 and so on. The map shows a relatively random
distribution of bonds. The scores of this map (25 variables)
were used as input of GA-PLS regression. The relative
importance of selected neurons for GA-PLS model is
shown in Figure 3s.
QSAR analysis
Analysis of empirical physicochemical parameters
by MOLMAP approach
The resulted GA-PLS models using MOLMAP scores as
input variables possessed high prediction ability to deter-
mine the activity class of compounds in the calibration set.
All molecules in this set (Table 1) were correctly predicted
as active, moderately active and inactive as it was experi-
mentally observed. The results are shown in Table 1. The
predictive ability of the model was measured by applica-
tion of 6 external test set molecules. The model also is able
to correctly predict the activity class of compounds. Five
out of six molecules were correctly predicted while mole-
cule 27 was experimentally active but was predicted to be
moderately active by this model. This model was applied
for the prediction of the activity class of 302 novel
3-hydroxypyridine-4-one derivatives. The structures of
these compounds are depicted in Table 1s. Most of these
compounds have not been synthesized and none of them
In the case of MOLMAP analysis, the empirical physico-
chemical parameters should be arranged in a three-way
array in the direction of molecules, bonds and empirical
physicochemical parameters. Then, this 3D array should be
introduced as input of Kohonen network to obtain the
related scores. Here, we investigated different dimensions
in the range of (4 9 4) to (13 9 13) over different epochs
(20–120) and, in each case; the resulting scores were used
as input of GA-PLS model. Best results were obtained by a
(5 9 5) over 20 epochs dimension for all data sets. The
PLS estimate of the regression coefficients is shown in
Figure 2s. The distribution of bonds in the resulted Ko-
honen map (5 9 5) is represented in Figure 3s. It should be
noted that the map contains 25 nodes, which can be
numbered sequentially from 1 to 25 so that the nodes of the
123

J Comput Aided Mol Des (2012) 26:349–361
355
Table 2 Predicted classes of
No.
Table 4 Specificity (SP), sensitivity (Sn), and precision (P) for all
the predefined activity classes obtained in cross-validation
GA-PLS
CP-ANN
N-aryl-3-hydroxypyridine-4-
ones 5a–s
5a
2
2
3
3
1
3
1
2
2
2
2
3
2
2
3
2
2
2
2
3
2
3
3
1
2
1
2
2
2
2
3
2
2
3
2
3
2
2
Activity class
SP
Sn
P
5b
5c
5d
5e
5f
1
2
3
0.97
0.91
0.80
1
0.94
0.92
0.93
0.77
0.93
5g
5h
5i
activity classes. The top map of the calculated model
(7 9 7 neurons and 60 epochs) is shown in the Fig. 2. In
the top map with 49 neurons, each sample is labeled on the
basis of the Kohonen weight of variable 1, going from low
values (white) to high values (black). Since the model was
built with a toroidal boundary condition, each edge of the
Kohonen map has to be seen as connected with the oppo-
site ones. It was seen that variable 1 is more influential on
the molecules of activity class 2 than the molecules of the
class 1 and 3. Therefore, it is possible to inspect each
variable on the top map, at the same time. The profile of
Kohonen weights of one of the neurons where class 2
molecules are placed is shown in Figure 4s. It was seen that
molecules in this class are characterized by having high
values of variable 1 and 4, low values of variable 2, and
extremely low values of variable 3. It is not possible to
have a comprehensive insight into the relationships
between variables and molecules, since we can only plot
the weights of one variable at a time for all the neurons
(Fig. 2) or, on the contrary, the variable profile of one
neuron simultaneously (Figure 4s). The weights of Koho-
nen layer are arranged as a data matrix W with 49 rows and
4 columns, as explained before, and PCA was calculated.
In Figs. 3 and 4 the score and loading plots of the first two
components (explaining together 83 % of the total
5j
5k
5l
5m
5n
5o
5p
5q
5r
5s
have been evaluated for antimicrobial activity yet. The
prediction results are shown in Table 2s. The prediction
results of N-aryl-3-hydroxypyridine-4-ones 5a–s are shown
in Table 2. Tables 1s, 2s and Figures 2s, 3s are shown in
the supplementary materials.
Analysis of MOLMAP descriptors by CP-ANNs approach
GA-PLS method applied a regression-based approach to
predict the activity of the studied compounds. The bio-
logical activity class of discrete values can also be modeled
by classification-based methods. Here CP-ANN was used
as a classification method. CP-ANNs were trained to pre-
dict the activity class of the compounds on the basis of four
MOLMAP descriptors selected by GA. To find optimal CP-
ANN settings, several networks were evaluated by chang-
ing the number of neurons and training epochs. Settings
were then selected on the basis of optimization of a clas-
sification parameter, such as error rate, evaluated in cross
validation. In Table 3 part of the classification indices
calculated by the toolbox (non error rate and error rate) is
shown. In Table 4, the other classification indices (speci-
ficity, sensitivity, and precision) are shown for all modeled
3
2
3
3
3
23
3
3
3
2
3
3
3
3
1
2
1
2
2
3
3
Table 3 Non-error rate (NER) and error rate (ER) obtained in fitting
and cross-validation
3
1
2
2
3
3
3
3
2
Fitting
Cross-validation
Fig. 2 Kohonen top map with toroidal boundary condition, each
molecule is labeled on the basis of its class; neurons are colored on
the basis of the weight of variable 1
NER
ER
0.83
0.17
0.79
0.21
123

356
J Comput Aided Mol Des (2012) 26:349–361
score plot
0.6
0.4
0.2
0
-0.2
-0.4
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
PC 1 - EV = 51.66%
Fig. 5 Representation of the CP-ANN output layer, with neurons
predicted as active, partially active, and inactive (green, blue and
yellow color, respectively)
Fig. 3 Score plot of the first two principal components calculated on
the Kohonen weights. Each neuron is colored on the basis of the
output weight of each activity class. Active, partially active, and
inactive neurons are in green, blue and yellow color, respectively
highlights the relationships between structural features and
the class of compounds. Inspection of a CP-ANN trained
with all molecules (Fig. 5) revealed that structurally simi-
lar molecules were mapped together in one neuron or as a
cluster of neurons. In this figure, the neurons with a low
output (inactive molecules) are colored in yellow and
neurons corresponding to active and moderately active
molecules are colored in green and blue, respectively.
Molecules 27, 29 and 30 which were predicted to be potent
appeared in a single neuron at the bottom left. Two of these
three molecules were experimentally observed to be active
(the first class) and one inactive (molecule 29 in the third
class). Molecules 1, 7, 10, 15, 25, 26 and 28 were clustered
to each other at the mid, bottom and top of the CP-ANN
surface in Fig. 5. All of these compounds were correctly
predicted to be moderately active.
information) are shown, respectively. In the score plot of
Figs. 3, each point represents a neuron of the previous CP-
ANN model. Each neuron is colored on the basis of the
output weight of each activity class (three classes). For
example the neurons assigned to class 2 are all clustered
and placed at the left side of the score plot. Variables 1 and
4 are placed at the left side of the loading plot and thus are
directly correlated with class 2; on the contrary, variable 3
has the largest positive loadings in the first component and
the molecules of class 2 will be characterized by small
value of this variable. The same conclusion could be made
by analyzing the single profile of Figure 4s. Figure 5
loading plot
0.8
Twenty-one molecules predicted inactive were clustered
into yellow neurons. Eighteen out of twenty-one molecules
were experimentally observed to be inactive whereas two
molecules were identified as moderately active (molecules
19 and 20) and one as active (molecule 22). Figure 4s is
shown in the supplementary materials.
1
0.6
0.4
2
0.2
0
3
Proposing new compounds
-0.2
-0.4
-0.6
-0.8
Based on the developed QSAR models by GA-PLS and
CP-ANN, and considering the structural similarities with
the studied compounds, 302 more compounds, most of
them have not been previously synthesized, were proposed
and their antimicrobial activities were predicted. The
results are shown in Table 2s and Table 2.
4
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
PC 1 - EV = 51.66%
To evaluate the predictability and accuracy of the
obtained QSAR models, 19 compounds belonging to all
three activity classes were selected among the 302
Fig. 4 Loading plot of the first two principal components calculated
on the Kohonen weights. Each variable is labeled with its identifi-
cation number
123

J Comput Aided Mol Des (2012) 26:349–361
357
3-[3-Hyroxy-2-methy-4-oxopyridin-1(4H)-yl] benzoic acid
(4-chloro-phenyl methylene)-hydrazide (5b) 118.87 mg
(82 %) compound was obtained as a light pink crystals,
m.p. 219–220 °C; FT-IR (KBr Disc) m/cm-1: 3150–3300
(broad, NH and OH), 3064 (CH, aromatic), 2925 (CH,
aliphatic), 1668 (C=O, ketone), 1628 (C=O, acylhydraz-
one), 1577 (shouldered C=C, C=N), 1491 (C=C aromatic);
1H-NMR (DMSO-d6, 400 MHz): (dppm) 12.00 (s, 1H,
NH), 8.42 (1H, N=CH17), 8.09 (bs, 1H, H10), 7.98 (bs, 1H,
compounds. The proposed compounds were synthesized
and their antimicrobial activity was evaluated against S.
aureus. Comparing the experimental results (as MIC) for
anti S. aureus activity and the predicted activity classes, the
accuracy of the obtained models can be judged. The col-
lected results in Table 2 disclose this fact that seventeen
out of the nineteen synthesized molecules were correctly
predicted by GA-PLS model according to antimicrobial
evaluation method. Two molecules (5f and 5h) proved to
be experimentally moderately active and active, respec-
tively but were predicted to be inactive and moderately
active by this model, respectively.
0
H13), 7.70–7.80 (m, 4H, H9, H11, H3 , H5 ), 7.62 (m, 1H,
0
0
0
H2), 7.52 (d, J = 8.00 Hz, 2H, H2 , H6 ), 6.23 (m, 1H, H3),
3.41(bs, 1H, OH), 1.99 (s, 3H,C5–CH3). MS (EI):
m/z = 199.2 [M?ꢀ–CONHN=CHC6H4Cl, 80 %], 381.2
The CP-ANN based prediction was correct for sixteen
out of the nineteen synthesized molecules. Three molecules
(5a, 5h and 5q) were moderately active and active based on
the antimicrobial assays but they were introduced as
members of inactive, moderately active and inactive class
of compounds according to CP-ANN model.
[M?ꢀ 35.5Cl), 100 %], 383.2 [M?ꢀ 37.5Cl), 38 %].
( (
3-[3-Hyroxy-2-methy-4-oxopyridin-1(4H)-yl] benzoic acid
(3-bromo-phenyl methylene)-hydrazide (5c) 121.69 mg
(75 %) compound was obtained as a light pink crystals,
m.p. 258–259 °C; FT-IR (KBr Disc) m/cm-1: 3433 (NH),
3359 (OH), 3062 (CH, aromatic), 2970 (CH, aliphatic),
1660 (C=O, ketone), 1628 (C=O, acylhydrazone), 1572
Chemistry
1
(shouldered, C=N, C=C), 1495 (C=C aromatic); H-NMR
Formation of the desired compounds was confirmed with
1
FT-IR, H-NMR and Mass spectra. All spectral data are
(DMSO-d6, 400 MHz): (dppm) 12.06 (s, 1H, NH), 8.40
(1H, N=CH17), 8.06 (s, 1H, H10), 7.94–7.98 (m, 2H, H13,
provided below. Carbon atoms are numbered sequentially
1
to facilitate the assignment of protons in H-NMR spectra
0
H2 ), 7.61–7.77 (m, 5H, H2, H9, H11, H4 , H6 ), 7.43 (t, 1H,
0
0
0
H5 ), 6.25 (d, J = 8.00 Hz, 1H, H3), 3.42 (s, 1H, OH), 2.00
(Scheme 1). The hydrazone N=CH- and -NH-signals
appeared as two singlets at 8.09–9.00 and 11.59–12.35,
respectively. H2 signal was seen among a group of signals
belonging to the aromatic protons. H3 was observed in all
(s, 3H,C5–CH3). MS (EI): m/z = 199.1 [M?ꢀ–CON-
HN=CHC6H4Br, 100 %], 427.2 [M?ꢀ (79Br)ꢀ, 66 %], 429.2
[M?ꢀ (81Br)ꢀ, 66 %].
3-[3-Hyroxy-2-methy-4-oxopyridin-1(4H)-yl] benzoic acid
(4-bromo-phenyl methylene)-hydrazide (5d) 113.58 mg
(70 %) compound was obtained as a light pink crystals,
m.p. 174–175 °C; FT-IR (KBr Disc) m/cm-1: 3100–3250
(broad, NH and OH), 3003 (CH, aromatic), 2925 (CH,
Aliphatic), 1668 (C=O, ketone), 1630 (C=O, acylhydraz-
one), 1570 (shouldered, C=N, C=C), 1487 (C=C aromatic);
1H-NMR (DMSO-d6, 400 MHz): (dppm) 11.98 (s, 1H,
NH), 8.40 (1H, N=CH17), 8.06 (bs, 1H, H10), 7.96 (bs,1H,
molecules as
a doublet (J constant = 8.00 Hz) at
6.22–7.20. Singlet signal belonging C5–CH3 hydrogens
was found in the range of 1.96–2.06. A broad singlet for
C4–OH appeared at 3.40–3.80. In all Mass spectra a frag-
ment with the molecular weight of about 199 was detected.
This fragment can be attributed to C12H9NO2 in which the
acylhydrazone moiety is cleaved and the hydroxyl proton
on the 3-hydroxyopyridin-4-one scaffold is gone.
3-[3-Hyroxy-2-methy-4-oxopyridin-1(4H)-yl] benzoic acid
(2-chloro-phenyl methylene)-hydrazide (5a) 118.87 mg
(82 %) compound was obtained as light pink crystals, m.p.
158–159 °C; FT-IR (KBr Disc) m/cm-1: 3130–3300 (broad,
NH and OH), 3064 (CH, aromatic), 2980 (CH, aliphatic),
1684 (C=O, ketone), 1628 (C=O, acylhydrazone), 1577
0
0
0
0
H13), 7.62–7.78 (m, 7H, H2, H9, H11, H2 , H3 , H5 , H6 ),
6.25 (bs, 1H, H3), 3.42 (bs, 1H, OH), 1.98 (s, 3H,C5–CH3).
MS (EI): m/z = 199.1 [M?ꢀ–CONHN=CHC6H4Br,
100 %], 427.2 [M?ꢀ (79Br), 83 %], 429.2 [M?ꢀ (81Br),
83 %].
1
3-[3-Hyroxy-2-methy-4-oxopyridin-1(4H)-yl] benzoic acid
(4-nitro-phenyl methylene)-hydrazide (5e) 119.16 mg
(80 %) compound was obtained as a light brown crystals,
m.p. 290–291 °C; FT-IR (KBr Disc) m/cm-1: 3329 (NH),
3211(OH), 3064 (CH, aromatic), 2925 (CH, aliphatic),
1672 (C=O, ketone), 1630 (C=O, acylhydrazone), 1570
(shouldered, C=N, C=C), 1489 (C=C aromatic); H-NMR
(DMSO-d6, 400 MHz): (dppm) 12.14 (s, 1H, NH), 8.84
(1H, N=CH17), 8.09 (bs, 1H, H10), 7.98–8.06 (m, 2H, H13,
0
H3 ), 7.70–7.76 (m, 2H, H9, H11), 7.65 (d, J = 7.20 Hz, 1H,
0
0
0
H2), 7.54 (d, 1H, H6 ), 7.42–7.48 (m, 2H, H4 , H5 ), 6.27 (d,
J = 7.20 Hz, 1H, H3), 3.42 (bs, 1H, OH), 2.00 (s, 3H, C5–
CH3). MS (EI): m/z = 199.1 [M?ꢀ–CONHN=CHC6H4Cl,
100 %], 381.2 [M?ꢀ
1
(shouldered, C=N, C=C), 1497 (C=C aromatic); H-NMR
( (
35.5Cl), 96 %], 383.2 [M?ꢀ 37.5Cl),
(DMSO-d6, 400 MHz): (dppm) 12.21 (s, 1H, NH), 8.54
0
(1H, N=CH17), 8.31 (d, J = 8.00 Hz, 2H, H3 , H5 ), 8.09 (s,
0
37 %].
123

358
J Comput Aided Mol Des (2012) 26:349–361
0
1H, H10), 8.00 (d, 3H, H9, H11, H13), 7.72–7.76 (d, 2H, H2 ,
(5i) 109.85 mg (73 %) compound was obtained as a yel-
:
low crystals, m.p. 284–285 °C; FT-IR (KBr Disc) m/cm-1
0
H6 ), 7.65 (d, J = 7.78 Hz, 1H, H2), 6.27 (d, J = 7.78 Hz,
1H, H3), 3.43 (bs, 1H, OH), 2.00 (s, 3H, C5–CH3). MS (EI):
m/z = 199.1 [M?ꢀ–CONHN=CHC6H4NO2, 75 %], 392.2
[M?ꢀ, 100 %].
3100–3250 (broad, NH and OH), 3050 (CH, aromatic),
2964 (CH, aliphatic), 1684 (C=O, ketone), 1630 (C=O,
acylhydrazone), 1579 (shouldered, C=N, C=C), 1487
(C=C Aromatic); 1H-NMR (DMSO-d6, 400 MHz):
(dppm) 12.27 (s, 1H, NH), 8.45 (1H, N = CH17), 8.22 (s,
3-[3-Hyroxy-2-methy-4-oxopyridin-1(4H)-yl] benzoic acid
(4-methyl-phenyl methylene)-hydrazide (5f) 102.88 mg
(75 %) compound was obtained as a light pink crystals,
m.p. 244–245 °C; FT-IR (KBr Disc) m/cm-1: 3250–3300
(broad, NH and OH), 3075 (CH, aromatic), 2922 (CH,
aliphatic), 1682 (C=O, ketone), 1630 (C=O, acylhydraz-
one), 1570 (shouldered, C=N, C=C), 1483 (C=C, aro-
0
1H, H4 ), 8.08 (bs, 1H, H10), 7.97 (bs, 1H, H13), 7.73–7.78
(m, 2H, H9, H11), 7.63 (d, 1H, H2), 6.24 (d, 1H, H3), 4.28
(s, 3H, N–CH3), 3.39 (bs, 1H, OH), 1.99 (s, 3H, C5–CH3).
MS (EI): m/z = 199.1 [M?ꢀ–CONHN=CHC3H4N3O2,
100 %], 396.2 [M?ꢀ, 33 %].
1
matic); H-NMR (DMSO-d6, 400 MHz): (dppm) 11.85 (s,
1H, NH), 8.39 (1H, N=CH17), 8.05 (bs, 1H, H10), 7.96 (bs,
3-[Hyroxy-2-methy-4-oxopyridin-1(4H)-yl] benzoic acid
[(5-nitrofuran-2-yl) methylene]-hydrazide (5j) 113.22 mg
(78 %) compound was obtained as a yellow crystals, m.p.
260–261 °C; FT-IR (KBr Disc) m/cm-1: 3100–3400 (broad,
NH and OH), 3066 (CH, aromatic), 2990 (CH, aliphatic),
1695 (C=O, ketone), 1635 (C=O, acylhydrazone), 1564
1H, H13), 7.71 (bs, 2H, H9, H11), 7.60–7.66 (m, 3H, H2,
0
H3 , H5 ), 7.28 (d, 2H, H2 , H6 ), 6.25 (d, J = 8.00 Hz, 1H,
0
0
0
0
H3), 3.42 (bs, 1H, OH), 2.34 (s, 3H, C4 –CH3), 1.99 (s,
3H,C5–CH3). MS (EI): m/z = 199.1 [M?ꢀ–CON-
HN=CHC6H4CH3, 75 %], 361.2 [M?ꢀ, 100 %].
(shouldered, C=N, C=C), 1477 (C=C aromatic); H-NMR
1
(DMSO-d6, 400 MHz): (dppm) 12.35 (s, 1H, NH), 8.40 (1H,
N = CH17), 8.09 (bs, 1H, H10), 7.99 (s, 1H, H13), 7.80 (d, 1H,
3-[3-Hyroxy-2-methy-4-oxopyridin-1(4H)-yl] benzoic acid
(2-methoxy-phenyl methylene)-hydrazide (5g) 117.47 mg
(82 %) compound was obtained as a white crystals, m.p.
187–188 °C; FT-IR (KBr Disc) m/cm-1: 3413 (NH), 3209
(OH), 3070 (CH, aromatic), 2927 (CH, aliphatic), 1653
(C=O, ketone), 1620 (C=O, acylhydrazone), 1599 (C=C,
aromatic), 1572 (shouldered, C=N, C=C), 1483 (C=C,
0
H4 ), 7.75–7.78 (m, 2H, H9, H11), 7.70 (d, J = 7.78 Hz, 1H,
0
H2), 7.30 (bs, 1H, H3 ), 6.34 (d, J = 7.78 Hz, 1H, H3), 3.40
(bs, 1H, OH), 2.01 (s, 3H, C5–CH3). MS (EI): m/z = 199.1
[M?ꢀ–CONHN=CHC4H2NO3, 100 %], 382.2 [M?ꢀ, 33 %].
3-[3-Hydroxy-2-methyl-4-oxopyridin-1(4H)-yl] benzoic
acid [(1-methyl-2-(methylthio)-1H-imidazol-5-yl)methy-
1
aromatic); H-NMR (DMSO-d6, 400 MHz): (dppm) 11.89
(s, 1H, NH), 8.79 (1H, N = CH17), 8.07 (s, 1H, H10), 7.98
lene]-hydrazide (5k) 98.05 mg (65 %) compound was
obtained as a yellow crystals, m.p. 278–279 °C; FT-IR
(KBr Disc) m/cm-1: 3100–3300 (broad, NH and OH), 3060
(CH, aromatic), 2933 (CH, aliphatic), 1674 (C=O, ketone),
1628 (C=O, acylhydrazone), 1574 (shouldered, C=N,
C=C), 1485 (C=C aromatic); 1H-NMR (DMSO-d6,
400 MHz): (dppm) 11.76 (s, 1H, NH), 8.33 (1H, N =
CH17), 8.03 (bs, 1H, H10), 7.93 (s,1H, H13), 7.68–7.72 (m,
2H, H9, H11), 7.63 (d, J = 8.00 Hz, 1H, H2), 7.41 (s, 1H,
0
(s, 1H, H13), 7.88 (d, J = 8.00 Hz, 1H, H6 ), 7.73–7.69 (m,
2H, H9, H11), 7.59–7.62 (bs, 1H, H2), 7.43 (t, J = 8.00 Hz,
0
0
0
1H, H5 ), 7.12 (d, J = 8.00 Hz, 1H, H3 ), 7.03 (t, 1H, H4 ),
6.24 (d, J = 7.78 Hz, 1H, H3), 3.37 (bs, 1H, OH), 3.86 (s,
3H, OCH3), 2.00 (s, 3H,C5–CH3). MS (EI): m/z = 199.2
[M?ꢀ–CONHN=CHC6H4OCH3, 90 %], 377.3 [M?ꢀ,
100 %].
3-[3-Hyroxy-2-methy-4-oxopyridin-1(4H)-yl] benzoic acid
(4-methoxy-phenyl methylene)-hydrazide (5h) 114.60 mg
(80 %) compound was obtained as a white crystals, m.p.
281–282 °C; FT-IR (KBr Disc) m/cm-1: 3200–3300 (broad,
NH and OH), 3068 (CH, aromatic), 2920 (CH, aliphatic),
1670 (C=O, ketone), 1628 (C=O, acylhydrazone), 1608
(C=C, aromatic), 1572 (shouldered, C=N, C=C), 1495
(C=C, aromatic); 1H-NMR (DMSO-d6, 400 MHz): (dppm)
11.79 (s, 1H, NH), 8.38 (1H, N = CH17), 8.05 (bs, 1H,
0
H4 ), 6.24 (d, J = 8.00 Hz, 1H, H3), 3.83 (s, 3H, N–CH3),
3.42 (bs, 1H, OH), 2.59 (s, 3H, S–CH3), 1.99 (s, 3H,
[M?ꢀ–CON-
C5–CH3).
MS
(EI):
m/z = 199.1
HN=CHC4H7N2S, 100 %], 397.2 [M?ꢀ, 96 %].
3-[3-Hydroxy-2-methyl-4-oxopyridin-1(4H)-yl
benzoic
acid (Phenylvinyl) -hydrazide. (5l) 110.55 mg (78 %)
compound was obtained as a yellow to white crystals, m.p.
189–190 °C; FT-IR (KBr Disc) m/cm-1: 3200–3300 (broad,
NH and OH), 3037 (CH, aromatic), 2935 (CH, aliphatic),
1674 (C=O, ketone), 1626 (C=O, acylhydrazone), 1562
0
H10), 7.96 (bs, 1H, H13), 7.65–7.73 (m, 4H, H9, H11, H2 ,
0
6 ), 7.63 (d, J = 7.2 Hz, 1H, H2), 7.02 (d, J = 8.00 Hz, 2H,
0
0
H3 , H5 ), 6.24 (d, J = 7.2 Hz, 1H, H3), 3.80 (s, 3H, OCH3),
1.99 (s, 3H, C5–CH3). MS (EI): m/z = 199.2 [M?ꢀ–
CONHN=CHC6H4OCH3, 45 %], 377.3 [M?ꢀ, 100 %].
1
(shouldered, C=N, C=C), 1479 (C=C aromatic); H-NMR
(DMSO-d6, 400 MHz): (dppm) 11.81 (s, 1H, NH), 8.30
(1H, N = CH17), 8.22 (bs, 1H, H10), 7.95 (bs, 1H, H13),
0
7.68–7.73(m, 2H, H9, H11), 7.60–7.66 (m, 3H, H2, H4 ,
3-[3-Hydroxy-2-methyl-4-oxopyridin-1(4H)-yl] benzoic acid
[(1-methyl-5-nitro-1H-imidazol-2-yl)methylene]- hydrazide
0 0 0 0
8 ), 7.31–7.40 (m, 3H, H5 , H6 , H7 ), 7.04–7.10 (m, 2H,
H
123

J Comput Aided Mol Des (2012) 26:349–361
359
0
0
H1 , H2 ), 6.24 (m, 1H, H3), 3.56 (bs, 1H, OH), 1.99 (s, 3H,
C5–CH3). MS (EI): m/z = 199.2 [M?ꢀ–CONHN=CHC8H7,
60 %], 373.3 [M?ꢀ, 100 %].
3-[3-Hydroxy-2-methyl-4-oxopyridin-1(4H)-yl]
benzoic
acid (3-nitro-phenyl methylen)-hydrazide (5s) Spectral
data for the compounds 5p–s are described elsewhere [32].
3-[3-Hyroxy-2-methy-4-oxopyridin-1(4H)-yl]) benzoic acid (4-
dimethylamino- phenyl methylene)-hydrazide (5m) 121.52 mg
(82 %) compound was obtained as a yellow crystals, m.p.
Antimicrobial activity
Minimum inhibitory concentrations (MICs) were deter-
mined by microdilution method. S. aureus was obtained
from the Persian type culture collection (PTCC), Tehran,
Iran (PTCC 1112). Mueller–Hinton agar was used to cul-
ture this microorganism for 24 h at 37 °C. The antibacte-
rial activity data are given in Table 5.
201–202 °C; FT-IR:(KBr Disc) FT IR: (KBr Disc) m/cm-1
:
3100–3500 (broad, NH and OH), 3072 (CH, aromatic), 2922
(CH, aliphatic), 1653 (C=O, ketone), 1630 (C=O, acylhyd-
razide), 1597 (C=N), 1576 (C=C), 1483 (C=C aromatic); 1H-
NMR (DMSO-d6, 400 MHz): (dppm) 11.61 (s, 1H, NH),
8.28 (1H, N=CH17), 8.04 (bs, 1H, H10), 7.94 (bs, 1H, H13),
7.68–7.72 (m, 2H, H9, H11), 7.62 (d, J = 8.00 Hz, 1H, H2),
Significant antibacterial activity was observed in some
compounds against S. aureus. Compounds 5e, 5g and 5h
inhibited the growth of this microorganism at 32 lg/mL.
Compounds 5a, 5b, 5k, 5n, 5p, 5r, 5s showed good
inhibitory activity (64 lg/mL) against its growth. It is
generally accepted that iron chelators inhibit microbial
growth by reducing iron absorption by microorganism.
0
7.53 (d, J = 8.00 Hz, 2H, H3 , H5 ), 6.77 (d, J = 8.00 Hz,
0
0
0
2H, H2 , H6 ), 6.23 (d, J = 8.00 Hz, 1H, H3), 3.42 (bs, 1H,
OH), 2.97 (s, 6H, N–CH3), 1.99 (s, 3H,C5–CH3). MS (EI):
m/z = 199.1 [M?ꢀ–CONHN=CHC8H10N, 25 %], 375.3
[M?ꢀ–CH3, 100 %], 390.3 [M?ꢀ, 100 %].
3-[3-Hydroxy-2-methyl-4-oxopyridin-1(4H)-yl] benzoic acid
(2-hydroxy-phenyl methylene)-hydrazide (5n) 93.79 mg
(68 %) compound was obtained as a yellow to white crystals,
m.p. 284–285 °C; FT-IR (KBr Disc) 3100–3300 (broad, NH
and OH), 3064 (CH, aromatic), 2925 (CH, aliphatic), 1670
(C=O, ketone), 1628 (C=O, acylhydrazide), 1570 (shoul-
dered, C=N, C=C), 1493(C=C aromatic); 1H-NMR (DMSO-
Table 5 In vitro antibacterial activity (MIC values, lg mL-1) of
different 3-hydroxypyridine-4-one derivatives
0
d6, 400 MHz): (dppm) 12.13 (s, 1H, NH), 11.12 (bs, 1H, C2 –
OH), 9.00 (1H, N = CH17), 8.64(bs, 1H, H10), 8.13 (bs,1H,
Compound
MIC (lg mL-1)
0
H6 ), 7.99 (s 1H, H13), 7.54–7.78 (m, 4H, H2, H9, H11, H3 ),
0
5a
64
64
0
0
7.39 (t, 1H, H4 ), 7.30 (t, 1H, H5 ), 6.25 (d, 1H, H3), 3.42 (bs,
5b
1H, OH), 1.99 (s, 3H,C5–CH3). MS (EI): m/z = 199.1 [M?ꢀ–
5c
256
256
32
CONHN=CHC6H4OH,
67 %],
227.1
[M?ꢀ
-
5d
NHN = CHC6H4OH, 100 %], 363.1 [M?ꢀ, 13 %].
5e
5f
128
32
3-[3-Hyroxy-2-methy-4-oxopyridin-1(4H)-yl] benzoic acid
ethylidene-hydrazide (5o)] 82.52 mg (75 %) compound
was obtained as a white crystals, m.p. 223–224 °C; FT-IR
(KBr Disc) m/cm-1: 3120–3300 (broad,NH and OH), 3062
(CH, aromatic), 2927 (CH, aliphatic), 1670 (C=O, ketone),
1631 (C=O, acylhydrazide), 1564 (shouldered, C=N, C=C),
1485 (C=C aromatic); 1H-NMR (DMSO-d6, 400 MHz):
(dppm) 11.59 (s, 1H, NH), 8.00–8.09 (m, 2H, N = CH17,
H10), 7.90 (bs, 1H, H13), 7.68–7.78 (bs, 2H, H9, H11), 7.60
(d, 1H, H2), 6.22 (d, 1H, H3), 3.49 (bs, 1H, OH), 1.97 (s,
5g
5h
32
5i
128
128
64
5j
5k
5l
256
128
64
5m
5n
5o
256
64
0
3H, C5–CH3), 1.86 (s, 3H, H1 ). MS (EI): m/z = 199.1
5p
[M?ꢀ–CONHN=CHCH3, 100 %], 285.1 [M?ꢀ, 7 %].
5q
128
64
5r
3-[3-Hydroxy-2-methyl-4-oxopyridine-1(4H)-yl] benzoic acid
[(furan-2-yl) methylene]-hydrazide (5p)
5s
64
Ampicillin
Tetracycline
256
64
3-[3-Hydroxy-2-methyl-4-oxopyridine-1(4H)-yl] benzoic acid
[(thiophene-2-yl) methylene]-hydrazide (5q)
MIC (lg/mL) = minimum inhibitory concentration, i.e., the lowest
concentration to completely inhibit bacterial growth
3-[3-Hydroxy-2-methyl-4-oxopyridin-1(4H)-yl] benzoic acid-
phenylmethylene-hydrazide (5r)
Ampicillin, Tetracycline: standard drug for bacteria
123

360
J Comput Aided Mol Des (2012) 26:349–361
8. Kotani T, Ichimoto I, TatsumI C, Fujita T (1975) Structure-
activity study of bacteriostatic kojic acid analogs. Agric Biol
Chem 39:1311–1317
9. Kotani T, Ichimoto I, Tatsumi C, Fujita T (1976) Bacteriostatic
activities and metal chelation of kojic acid analogs. Agric Biol
Chem 40:765–770
10. Feng M, van der Does L, Bantjes A (1993) Iron (III)-chelating
resins. 3. Synthesis, iron (III)-chelating properties, and in vitro
antibacterial activity of compounds containing 3-hydroxy-2-
methyl-4 (1H)-pyridinone ligands. J Med Chem 36:2822–2827
11. Fassihi A, Abedi D, Saghaie L, Sabet R, Fazeli H, Bostaki Gh,
Deilami O, Sadinpour H (2009) Synthesis, antimicrobial evaluation
and QSAR study of some 3-hydroxypyridine-4-one and 3-hydrox-
ypyran-4-one derivatives. Eur J Med Chem 44:2145–2157
12. Hansch C, Hoekman D, Gao H (1996) Comparative QSAR:
toward a deeper understanding of chemicobiological interactions.
Chem Rev 96:1045–1075
13. Horvath D, Mao B (2003) Neighborhood behavior. Fuzzy molec-
ular descriptors and their influence on the relationship between
structural similarity and property similarity. QSAR Comb Sci
22:498–509
14. Putta S, Eksterowicz J, Lemmen C, Stanton R (2003) A novel
subshape molecular descriptor. J Chem Inf Comput Sci 43:
1623–1635
15. Ballabio D, Consonni V, Todeschini R (2007) Classification of
multiway analytical data based on MOLMAP approach. Anal
Chim Acta 605:134–146
16. Zhang Q, Aires-de-Sousa J (2007) Random forest prediction of
mutagenicity from empirical physicochemical descriptors.
J Chem Inf Model 47:1–8
Conclusion
QSAR models allow medicinal chemists to predict the
biological activities of untested and even not prepared
compounds. Using traditional techniques, it may take
months to synthesize a new compound for biological
assays. Application of computational techniques for
designing biologically active novel compounds reduces
experimental research cost and saves a lot of time. Search for
new antimicrobial agents with novel modes of action rep-
resents a major target in anti infective chemotherapy. The
reason is the increasing number of pathogenic bacteria and
fungi that are resistant to therapeutic agents. 3-Hydroxy-
pyridine-4-one derivatives have shown good inhibitory
activity against bacterial strains such as S. aureus. In this
work we reported the application of MOLMAP descriptors
based on empirical physicochemical properties with
GA-PLS and CP-ANN methods to predict some novel
3-hydroxypyridine-4-one derivatives with improved anti-
bacterial activity against this microorganism. A large col-
lection of 302 novel derivatives of this chemical scaffold was
selected for this purpose, which the activity class of these
compounds was determined using the two QSAR models. To
evaluate the predictability and accuracy of the obtained
models, nineteen compounds belonging to all three activity
classes were prepared and the anti S. aureus activity of them
was determined. Comparing the experimental results (as
MICs) and the predicted activity classes revealed the accu-
racy of the GA-PLS and CP-ANN models.
17. Gupta S, Matthew S, Abreu P, Aires-de-Sousa J (2006) QSAR
analysis of phenolic antioxidants using MOLMAP descriptors of
local properties. Bioorg Med Chem 14:1199–1206
18. Zupan J, Gasteiger J (1999) Neural networks in chemistry and
drug design, 2nd edn. Wiley-VCH, Weinheim
19. Simon V, Gasteiger J, Zupan J (1993) A combined application of
2 different neural-network types for the prediction of chemical-
reactivity. J Am Chem Soc 115:9148–9159
20. Program PETRA can be tested on the website http://www2.
chemie.uni-erlangen.de/software/petra and is developed by Molecu-
lar Networks GmbH,http:www.mol-net.de
In sum, it seems that the MOLMAP approach is a
powerful and reliable method to assist medicinal chemists
for a rational design of successful bioactive agents.
21. Zhang Q, Aires-de-Sousa J (2005) Structure-based classification
of chemical reactions without assignment of reaction centers.
J Chem Inf Model 45:1775–1783
References
22. Sabet R, Fassihi
A (2008) QSAR study of antimicrobial
3-hydroxypyridine-4-one and 3-hydroxypyran-4-one derivatives
using different chemometric tools. Int J Mol Sci 9:2407–2423
23. Loncle C, Brunel J, Vidal N, Dherbomez M, Letourneux Y
(2004) Synthesis and antifungal activity of cholesterol-hydrazone
derivatives. Eur J Med Chem 39:1067–1071
24. Melnyk P, Leroux V, Sergheraert C, Grellier P (2006) Design,
synthesis and in vitro antimalarial activity of an acylhydrazone
library. Bioorg Med Chem Lett 16:31–35
´
1. Arredondo M, Nun˜ez MT (2005) Iron and copper metabolism.
Mol Asp Med 26:313–327
2. Skaar EP, Gaspar A, Schneewind O (2006) Bacillus anthracis
IsdG,
1071–1080
a heme-degrading monooxygenase. J Bacteriol 188:
3. Hider R, Kong X (2010) Chemistry and biology of siderophores.
Nat Prod Rep 27:637–657
4. Aytemir M, Erol D, Hider R, Ozalp M (2003) Synthesis and
evaluation of antimicrobial activity of new 3-hydroxy-6-methyl-
4-oxo-4H-pyran-2-carboxamide derivatives. Turk J Chem 27:
757–764
25. Kuecuekguezel S, Rollas S, Kuecuekguezel I, Kiraz M (1999)
Synthesis and antimycobacterial activity of some coupling
products from 4-aminobenzoic acid hydrazones. Eur J Med Chem
34:1093–1100
5. Aytemir M, Hider M, Erol D, Ozalp M, Ekizoglu M (2003)
Synthesis of new antimicrobial agents; amide derivatives of
pyranones and pyridinones. Turk J Chem 27:445–452
6. Qiu D, Huang Z, Zhou T, Shen C, Hider R (2011) In vitro
inhibition of bacterial growth by iron chelators. FEMS Microbiol
Lett 314:107–111
7. Dehkordi L, Liu Z, Hider R (2008) Basic 3-hydroxypyridin-4-
ones: potential antimalarial agents. Eur J Med Chem 43:
1035–1047
26. Dhar D, Taploo C (1982) Schiff bases and their applications. J Sci
Ind Res 41:501–504
27. Przybylski P, Huczynski A, Pyta K, Brzezinski B, Bartl F (2009)
Biological properties of schiff bases and azo derivatives of phe-
nols. Curr Org Chem 13:124–148
28. Todeschini R (2010) Milano Chemometrics and QSPR Group.
http://michem.disat.unimib.it/. Accessed 14 March 2010
29. Haaland D, Thomas E (1988) Partial least-squares methods for
spectral analyses. 1. Relation to other quantitative calibration
123

J Comput Aided Mol Des (2012) 26:349–361
361
methods and the extraction of qualitative information. Anal Chem
60:1193–1204
33. Kirilmis C, Ahmedzade M, Servi S, Koca M, Kizirgil A, Kazaz C
(2008) Synthesis and antimicrobial activity of some novel
derivatives of benzofuran: part 2. The synthesis and antimicrobial
activity of some novel 1-(1-benzofuran-2-yl)-2-mesitylethanone
derivatives. Eur J Med Chem 43:300–308
30. Kuzmanovski I, Novic M (2008) Counter-propagation neural
network in Matlab. Chem Intell Lab Sys 90:84–91
31. Harris RLN (1976) Potential wool growth inhibitors. Improved
syntheses of mimosine and related 4(1H)-pyridones. Aust J Chem
29:1329–1334
34. de Souza SM, Delle Monache F, Smania A Jr (2005) Antibac-
terial activity of coumarins. Z Naturforsch 60:693–700
32. Sabet R, Behjati M, Vahabpour R, Memarnejadian A, Rostami
M, Fassihi A, Aghasadeghi MR, Saghaie L, Miri R (in press) Int J
Cardiol. doi:10.1016/j.ijcard.2011.11.067
123