Professional Documents
Culture Documents
CBDD 12118
CBDD 12118
Research Article
attraction for the development of new vaccines (4). On the lidene hydrazine was identified as a lead. A set of phenyla-
other hand, because elimination of the parasites in the liver minoacetic acid benzylidene hydrazine analogs were then
prevents the initiation of the blood stage (BS) infection, designed and those found to bind tightly to pfENR were
drugs targeting the LS represent significant potential in taken up for synthesis. The structures of these molecules
causal prophylaxis of malaria. Primaquine is the only drug were confirmed by IR, 1H, 13C, 15N, 2D-NMR, and X-ray
specifically developed to inhibit the LS infection but suffers diffraction. Further, Recursive Partitioning (RP) analysis
from toxicity, poor compliance, and risk of hemolysis (5). was performed on these compounds to develop a classifi-
cation model based on their activity using physicochemi-
The liver phase has not been explored as a drug target, cal, structural, electronic and thermodynamic properties.
mainly due to technical challenges in studying this stage. The RP method is a specific case of the ‘binary QSAR’
Recent studies showed that de novo type II fatty acid bio- approach which can classify analog activity data through
synthesis (FAS-II) pathway, which was previously thought essential decisive elements that categorize the dataset into
to operate in BS, is only expressed in LS. Deletion of indi- groups of molecules based on the activity.
vidual FAS-II enzymes prevents the generation of erythro-
cytic merozoites, that is, the erythrocytic stage infection
cannot initiate (6,7). This data plus the fact that P. falcipa- Materials and Methods
rum employs a dissociated type of pathway (type II FAS)
for fatty acid synthesis, whereas humans use an associ- Virtual screening
ated pathway (type I FAS) renders the FAS-II pathway an Virtual Screening is an in silico approach that filters an
attractive target. Plasmodium falciparum enoyl-ACP reduc- extensive library of compounds (available or virtual) to a
tase (pfENR or FabI) (8,9) is a key enzyme in the type II limited number of potentially promising compounds for the
FAS pathway which reduces the trans-2 enoyl bond of target of interest by means of computational methods such
enoyl-ACP substrates to saturated acyl-ACPs and plays a as molecular docking. High-throughput molecular docking
decisive role in completing the fatty acid elongation cycles. involves screening a large database of molecules against
Gene knockout of pfENR from P. falciparum has shown the active site of the receptor followed by application of a
that this enzyme is critical for development of the parasite scoring function to estimate the possibility that the ligand
in the hepatic phase. Triclosan, [5-chloro-2-(2,4-dichloro- will bind to the protein with high binding affinity (15,16).
phenoxy)phenol], is a wide spectrum antifungal and anti- Molecular docking has a great advantage over 2D similarity
microbial agent which has been found to inhibit the fatty and 3D pharmacophore search techniques as it utilizes the
acid synthesis pathway (FAS ΙΙ) by binding to enoyl-ACP 3D structure of the receptor in a quantitative way.
reductase (10,11). It has a Ki of 50 nM against pfENR and
has high nanomolar to low micromolar IC50 against both The program Glide (17,18) (Grid-Based Ligand Docking
drug sensitive and multidrug resistant organisms. How- with Energetics) incorporated in the Schrodinger molecular
ever, due to its bioliability, the use of triclosan in humans modeling package (Schrodinger, Inc., USA) was used to
as an antimalarial drug is not yet suitable and several stud- perform molecular docking. It works on the principle of
ies have reported triclosan resistance in E. coli. In addition, systematic search for conformations, orientations and
triclosan presents poor bioavailability indicating a potential positions of the ligand in the receptor active site and elimi-
limitation in the use of this compound as a therapeutic nates unwanted conformations by a scoring function. This
agent. As a result, several attempts have been made to is followed by energy optimization.
modify triclosan (12,13) but have met with little success.
This underscores the need to develop novel strategies to Flexibility of the ligand/protein was considered during
overcome existing barriers. docking, allowing for a flip of 5- and 6-membered rings. A
maximum of 10 poses per ligand were stored and these
Our aim was to design agents that are more polar in char- were energy minimized postdocking. The receptor active
acter than triclosan and thus would not suffer from bioli- site was shaped by Tyr267, Gly313, Pro314, Ile323,
ability or bioavailability issues and more importantly agents Phe368, Ile369, Ala372, Tyr277, Ala319, and Ala320; resi-
with a same mechanism of action as triclosan, that is, an dues in a 10 A vicinity of triclosan. The molecules were
inhibitor of pf ENR. A detailed investigation of the X-ray docked into the active site using the Glide Extra-Precision
crystal structure of pf ENR in complex with triclosan (14) (XP) docking algorithm. Prediction of the binding affinity
could provide a clear understanding of the active site and and rank ordering of ligands in the database was achieved
the mode of interaction of triclosan with the target enzyme. by the Glide Score scoring function.
This information served as a guideline to initiate a rational
design approach to develop alternatives to triclosan that An important step in virtual ligand screening (VLS) is the
target the enoyl-ACP reductase (ENR) of P. falciparum. docking protocol that is adopted to retrieve relevant hits
With this objective, a virtual screen of drug-like-molecules from the database. The docking protocol should be able
from the iResearcha database, against the target (pfENR), to correctly predict the binding modes of the molecules in
was carried out using molecular docking. Based on the the database. The docking protocol adopted herein was
results of the virtual screen, phenylaminoacetic acid benzy- validated by reproducing the experimentally observed
Figure 1: (A) Overlay of the experimental binding mode of triclosan over best pose obtained by molecular docking into the crystal
structure (PDB ID: 1nhg). (B) Binding mode of molecule 6 obtained by molecular docking.
binding pose of triclosan in complexation with the target Table 1: Structures of the molecules designed as inhibitors of
pf ENR (PDB code 1NHG). It was observed that the top pfENR along with their docking scores and in vitro pf ENR inhibi-
10 docking solutions for triclosan were all within 1.0 A tory activity
RMSD of the crystal structure. The best scoring pose for R
triclosan obtained by docking has been compared with
the crystal structure in Figure 1A. A very good agreement H R'
N
was observed between the pose of the inhibitor found by N N
docking and the crystal structure. This assured us that the H
docking protocol would be good enough to correctly pre- O
dict the binding modes of the molecules in the database
during virtual screening. With the receptor grid generated
Mol. ID R R′ IC50 (lM) Glide score
for triclosan in the validation study, docking was performed
on drug-like molecules in the iResearch database compris- Triclosan 0.049 8.60
ing of 14 million compounds against the target (pf ENR) to 1 – o-Cl 90 7.75
search potential candidates. 2 – m,p-diOCH3 >159 6.75
3 p-Cl m,p-diOCH3 >143 6.94
4 p-Cl – 10 7.68
Chemistry 5 p-Cl o-Cl n.m 6.92
6 p-F o-OH 12 8.33
Synthesis was carried out using reported procedures (19
7 p-F p-N(CH3)2 8 7.34
–21) and reactions were modified to improve the yield 8 – – 197 7.53
and purity of the products. All solvents and reagents 9 – m-Br 90 6.90
used for synthesis were of general reagent grade. The 10 – o-OH >185 8.23
reactions were monitored by Thin Layer Chromatography 11 – p-N(CH3)2 >168 7.09
(TLC) with Merck precoated silica plates (GF 254) for 12 m-Cl, p-F o-OH >155 8.07
completion and for establishing the purity. Melting points 13 m-Cl, p-F – >163 7.35
14 m-Cl, p-F o-Cl 29 6.95
were recorded in open capillaries on an electrically
15 m-Cl, p-F m-Br >129 5.72
heated melting point apparatus and are uncorrected. IR
16 p-Cl o-OH >164 8.34
spectra were recorded (KBr disc method) on a Jasco FT- 17 p-Cl m-Br 7 7.26
IR 5300 spectrophotometer. NMR spectra were recorded 18 p-F o-Cl 163 6.58
on a Bruker AV500 instrument using DMSO-d6 as well 19 m-Cl – 173 7.03
as CDCl3 as solvents. 2D-NMR spectra were recorded 20 m-Cl o-Cl 77 6.31
with standard routines available on the spectrometer. 21a m-Cl m-Br 163 7.32
Chemical shifts are reported in ppm down field from
n.m., not measurable due to self-absorbance at 340 nm.
tetramethylsilane (TMS) as the internal standard. Micro- a
The activity reported for molecule 21 is the cyclic imidazole-4-
wave-assisted reactions were carried out using the one derivative (see Scheme 2 for structure and the text for more
Discover microwave synthesizer (CEM, Buckingham, details).
UK). The molecules (Table 1) were synthesized in three
steps as described below (Scheme 1). solid material obtained upon cooling was filtered and recrys-
tallized from hexane to yield ethyl phenylaminoacetate (iii).
R R
O
Cl 90°C
+ N
O
CH 3COONa
NH2 O H
O
(i) (ii) (iii)
NH2 NH 2
300 Watt/30 secs
R' R
H R'
N
N N 300 Watt/60 secs + H
N
H O
O N NH2
H
O
(vi) (v) (iv)
Scheme 1: Scheme for the synthesis of phenylaminoacetic acid benzylidene hydrazine analogs.
microwave synthesizer. After irradiation for 30 seconds binary splits for each descriptor and selects the optimal
(300 Watts), the product solidified on cooling to room tem- one. At each non-terminal node in the decision tree, the
perature. This was purified by recrystallization from hot dataset is split into two groups by a particular descriptor
ethanol to yield phenylaminoacetic acid hydrazide (iv). recursively until no further splitting is possible or a thresh-
old value of the termination rule is reached. The tree dis-
play of the SAR is clear, easy to interpret, and often
General procedure for the synthesis of suggestive. The RP models can be generated very fast
phenylaminoacetic acid benzylidene hydrazine and compared with other grouping techniques and the output
its derivatives (vi) provides a graphical tree, as well as a statistics table that
Compound (iv) (5 mmol) was mixed with benzaldehyde (v) includes correctly classified percentages, the number of
(5 mmol) in a microwave flask and the mixture was irradi- false positives, and compound member information for
ated in the Discover microwave synthesizer for 60 seconds each node. Recursive partitioning model was generated
(300 Watts). After completion of the reaction as indicated using the CERIUS2 (version 4.8; Accelrys Inc. USA)b soft-
by TLC, the solid product was recrystallized from ethanol ware. The detail procedure adopted for performing the PR
to give the pure phenylaminoacetic acid benzylidene has been discussed in the supporting information.
hydrazine (vi).
pounds was described in terms of their IC50 values, and docking scores are given in Table 1 and a representative
the dataset was split into two classes – less active (0) and docking pose is shown in Figure 1B.
active (1) based on these IC50 values. Class 1 (less active)
contains 13 compounds with IC50 values greater than It was observed that all 20 molecules adopted a binding
100 lM, while class 2 (active) contains 8 compounds mode which is complementary with the pf ENR active
with IC50 values lower than or equal to 100 lM. Prior to pocket. A very small difference in Glide score as well as
generating models, descriptors with zero variance as well a low RMSD observed upon superposition of the binding
as those containing 95% zero values were eliminated poses among all the 20 phenylamino acetic acid benzy-
because these are not useful for explaining the variation in lidene hydrazine analogs suggests an analogous binding
the activity. mode for all the candidates. The Glide docking score for
these analogs varied from 8.34 to 5.72 kcal/mol,
The correlation matrices for the descriptors were con- while triclosan has a score of 8.60 kcal/mol. This
structed and those descriptors with a cross-correlation means that these analogs could be developed as poten-
coefficient >0.5 were eliminated, because they contain tial candidates for second-generation drug development
nearly the same information. The uncorrelated descriptors against pfENR.
formed the independent variables (X), while the biological
activity served as the dependent variable (Y) for the analy-
sis. The CARTTM (classification and regression trees) NMR and conformational analysis
method was used to generate the decision tree. The activ- It is a well-known fact that small linear molecules are not
ity classes were weighted equally, and the tree was rigid in solution but have several degrees of internal
formed by optimizing the Twoing rule scoring function. The motional freedom. This results in several conformations
pruning factor was varied between 0 and 3. Also the val- being populated at room or physiological temperature. As
ues for maximum tree depth were varied from 5 to 10, a first step in understanding the conformational isomerism
while default values were used for maximum number of in phenylaminoacetic acid benzylidine hydrazides, a
generic splits (30), and the number of knots per variable detailed 1HNMR characterization was carried out. Phenyla-
(20). Internal validation was performed using cross-valida- minoacetic acid benzylidine hydrazides can in principle
tion, setting the number of cross-validation groups to 10. exist in four different conformations arising from combina-
tions of E and Z geometrical/conformational isomers at the
imine and the amide functional groups. These configura-
Results and Discussion tional and conformational isomers are depicted in Figure 3.
Design of substituted phenylamino acetic acid Of these structures, the E, Z and Z, Z configurations are
benzylidenehydrazines usually of higher energy, less stable and are not consid-
From among the several hits obtained from virtual screen- ered. The NMR spectra of these compounds in DMSO-d6
ing of drug-like molecules in the iResearch database, showed doubling of the signals indicating that only two of
ligand 45015 (Figure 2) was chosen as a candidate based these are present in solution. These two forms are likely to
on its similarity to triclosan and also on the characteristics be the Z, E and E, E isomers which differ in the position of
of the enzyme active site to bind this ligand specifically. the phenylaminoacyl group relative to the amide C-N
This hit was used as a starting point to design a series of
potential candidates.
Figure 2: The hit obtained by virtual screening of the iResearch Figure 3: Configurational and conformational isomers observed
database against the crystal structure of pfENR (PDB ID: 1nhg). in phenyl amino acetic acid benzylidene hydrazides.
bond. The assignments of the signals for the Z, E and E, E studied. Incidentally, the crystal structure obtained for mol-
isomers can be made on the basis of just 1H and 13C ecule 2 corresponds to the Z, E form (vide infra).
chemical shifts. In general, the azomethine protons and
carbons of the E, E isomer are known to be more shielded As mentioned earlier, molecule 21 was found to be differ-
compared with the Z, E isomer, while the amide carbonyl ent from the other members in the series. Elucidating the
shows the reverse trend, that is, they are shielded in the structure of molecule 21 was a difficult task, but never-
Z, E isomer compared with E, E isomer (27–29). theless was achieved using 1H-13C and 1H-15H-HMBC
along with NOESY spectra. A careful analysis of the 1H
In the present investigation, the major species present in and 13C NMR data indicated that the CH2 protons
DMSO solution corresponds to the E, E form. The protons appear (d 4.49 and 4.24) as an AB part of an ABX spin
and carbon atom of the CH2 group attached to the amide system with geminal coupling of ~15 Hz. Only one of the
function are also found to show considerable changes in CH2 protons shows very weak scalar coupling (2.7 Hz) to
the chemical shifts in these two forms (Tables 3 and 4). the X proton at d 6.15. The 13C chemical shifts of these
Further, the two forms present in solution are in dynamic carbons are 51.39 and 78.9, respectively. A deshielded
equilibrium. Evidence for this comes from the NOESY (Fig- azomethine proton (d 9.49) and four additional aromatic
ure 4) where strong positive exchange cross-peaks are protons are also seen in the proton spectrum. The struc-
seen between the major and minor signals of the same ture of this compound has been unambiguously eluci-
proton. The NOESY spectrum also indicated a weak dated by detailed 1D and 2D NMR investigations by 1H,
13
cross-peak between the CH2 protons and the ortho pro- C, and 15N techniques (Figure S1–S9) as an imidazoli-
tons of the Ar2 ring, indicating a favor for the E, E form. dine-4-one. 1H-13C HMBC correlations of the CH2 pro-
The high field CH2 carbons of the major E, E form show a tons, the CH proton at d 6.15, the azomethine proton
1
H-13C HMBC correlation to the low field NH (major) pro- and the CH carbon at d 78.9 were found to be very cru-
ton, whereas no correlation is seen from the minor CH2 to cial in arriving at the structure. For example, the CH car-
the minor NH under the conditions of measurement bon at d 78.9 shows correlations to the ortho protons of
(Figure 5). In the E, E form the 3JC-C-N-H is expected to be the bromophenyl group and also to the CH2 protons.
larger than that in the Z, E form. Thus, the predominance While the CH proton at d 6.15 shows connectivities to
of the E, E configuration can be envisaged in all the cases the carbon at d 164.56, the ipso carbons of the two aro-
matic rings (d 144.8 and 140.05) and to the two ortho
protons of one of the rings. The azomethine proton cor-
relations are seen mainly to the carbons of the third aro-
matic ring. The CH2 protons in turn show clear
Table 3: 1H and 13C chemical shifts distinguishing the major and
minor isomers of the imine group in different compounds
correlations to the CH carbon at d 78.9, the quaternary
carbons at d 164.56, 140.05, and 144.8. All these obser-
Imine (major) (d, Imine (minor) (d, vations could only be satisfied by considering a 1,2,3-tri-
ppm) ppm) substituted imidazolidine-4-one (Scheme 2). The HMBC
correlations are shown in Figures 6, 7 and S5–S7 along
Compound HC=N- HC=N-
with a schematic representation of the correlations
4 1
H 8.02 1
H 8.25 observed.
13 13
C 144.05 C 147.38
1 1
2 H 7.76 H 7.92 This product has been formed by reaction of the aldehyde
13 13
C 145.56 C 149.30 (3-bromobenzaldehyde) with the NH of the phenylaminoa-
1 1
7 H 7.89 H 8.08 cyl moiety (iii, Scheme 1) followed by cyclization through
13 13
C 144.90 C 148.25
1 1 dehydration (Scheme 2). A reaction of phenayaminoacyl
11 H 7.89 H 8.09
13
C 144.88 13
C 148.21
hydrazone with excess aromatic aldehyde (1:2) indeed
gave this product. The proposed structure is also sup-
Figure 4: NOESY spectrum of molecule 4 recorded at 298K (A) D1: 3.6–11.6 ppm, D2: 3.6 – 11.6 ppm; (B) D1: 3.74 – 4.34 ppm, D2:
5.8 –7.8 ppm.
Figure 5: (A) 13C HMBC spectrum of molecule 4 recorded at 298K; D1: 11.0–11.5 ppm, D2: 43.2–47.1 ppm (B) 15
N HMBC spectrum of
molecule 4 recorded at 298K; D1: 5.6–11.9 ppm, D2: 0–400 ppm.
ported by the 1H-15N HMBC (Figure 7) and the NOESY more stable E configuration. The nOe data are also in
correlations (Figures S9–S10). The most conclusive evi- agreement with this as no nOes are observed between the
dence for the NOESY data are the weak nOe’s between protons of the aromatic ring (7.53d, 7.27d, and 7.79d) on
the protons on the various rings. For example, the one the imine carbon and the CH2 group in the five-membered
between the ortho protons of the aromatic rings A and B ring. Besides, weak nOes are observed between the imine
and between the protons on the imidazolidine-4-one ring CH at ~9.5d and the inequivalent CH2 protons at 4.19 and
(D) with protons on rings A and B. 4.44d (Figure S12). During the formation of molecule 21,
an asymmetric center is created on the five membered
The 1H NMR spectrum of molecule 21 (Figure S11) shows ring. It very likely that the product formed is racemic in
only a single signal for the imine CH proton at ~9.5d, nature. We have not made any attempt to identify the
indicating the presence of only one geometrical isomer of chirality of the molecule because this molecule is found
the imine. Considering the bulkiness of the substituents, to be of less importance in the context of the studies
we assume this molecule will have a preference for the conducted.
H N
N N
C Br
O H
O Cl
O
Cl
H CH
N NH N
OH
Br
Br
Br
CH
N N
Cl N
Br
Scheme 2: Formation of
the imidazolidine 4-one derivative
Br (molecule 21).
belong to Orthorhombic, space group Pbca, a = 8.419 (1), H…O interactions. The hydrogen bond distances are N(2)-
b = 18.234 (2), c = 22.063(3) A, V = 3386.8(8)
A3, Z = 8, H…..O(3) = 2.836(3)A and N(3)-H(3)….O(1) = 3.084(3) A
Dc = 1.433 g/cc, l (MoKa) = 0.367 per mm, T = 293 (2) giving rise to a H-bonded helical structure as depicted in
K, 13003 reflections measured, 2965 unique [I > 2r(I)], R Figure 9 which is seen as an infinite one-dimensional chain
value 0.0492, xR2 = 0.1118. All the data were corrected (Base Vector: [1 0 0]). Analysis of potential hydrogen
for Lorentzian, polarization, and absorption effects. bonds is given in Table 5.
SHELX-97 (ShelxTL) (30) was used for structure solution
and full matrix least squares refinement on F2. Hydrogen
atoms were included in the refinement as per the riding pfENR inhibition activity (FabI Assay)
model. Data collection and refinement parameters are The pf ENR protein was cloned, expressed, and purified
listed in Table S1. by methods reported earlier (31,32). All measurements
were carried out on a Perkin Elmer Lambda 25 UV/VIS
Crystal data, structure refinement, atomic co-ordinates, spectrophotometer in a buffer solution containing 20 mM
bond lengths, bond angles and torsion angles anisotropic/ HEPES, pH 7.4 and 150 mM NaCl. Solutions for assay
isotropic displacement parameters are given in the sup- were prepared by dissolving compounds in DMSO (max.
porting information (Tables S2–S6). final concentration was 1%). For measuring pfENR inhibi-
tion, 1 lL of 100 lM NADH cofactor (43420; Sigma-
The ORTEP diagram of molecule 2 with the atom-number- Aldrich, St.Louis, MO, USA) was added to 1 lg of enzyme
ing scheme is displayed in Figure 8. The important bond and then 1 lL of varying compound concentration in a
lengths, bond angles, and torsion angles are given in total of 1 mL buffer. The reaction was started by adding
Tables S2–S6. As is evident from the table of torsion 1 lL of 50 lM crotonyl-CoA substrate (C6146; Sigma) and
angles, the backbone of the molecule is nearly extended the absorbance of the mixture was read spectrophotomet-
except for the torsion angle C11-N3-C12-C13 which rically for 1 min at 340 nm. For comparison purposes, tri-
describes the folding of the Ar2 ring in a gauche arrange- closan was chosen as the reference standard and it
ment. The observed bond parameters indicate that the exhibited an IC50 of 0.04 lm. IC50 values were estimated
molecule forms hydrogen bonded helical structure via N- from graphically plotted dose–response curves (Table 1).
H….O interactions (Figure 9). This is a unique attribute for
this class of compounds and to the best of our knowl- Eight of the 20 molecules were found to be active against
edge, helical structures of this type have not been pfENR with an IC50 value below 100 lm, while rest of the
reported earlier. There are two intermolecular H-bond N- analogs displayed moderate to low activity. These analogs
Table 5: Analysis of potential hydrogen bonds cyclized structure is not more potent than the linear phe-
nylaminoacetic acid benzylidene hydrazine scaffold, and
D-H….A D-H H….A D….A Angle
hence, no further attempts were made in exploring the
No (
A) (
A) (
A) (
A) D-H…A (o)
cyclized structure for the other molecules (1–20) in the
1 N(2)-H(2N)…O(3)i 0.86 2.04 2.836(3) 154 series.
2 N(3)-H(3)…O(1)ii 0.86 2.26 3.084 161
Table 6: ADME and toxicity prediction data of the phenylaminoacetic acid benzylidine hydrazine derivatives
Acute Acute
Physicochemical toxicity toxicity
Solubility determinants Distribution LD50 LD50
Genotoxicity Mouse Rat
Water Buffer (pH 7.4) logD Plasma protein
Mol. ID (logSw) (logS) (pH 7.4) logP binding Vd (L/kg) Ames test Oral mg/kg Oral mg/kg
Table 7: Statistical results of recursive partitioning Table 8: Activity prediction by recursive partitioning (PR)
the statistical significance of the RP model for classifying a Table 9: Cross-validation statistics of the recursive partitioning
new library of compounds. model
Class% Overall%
Figure 10 shows the optimized 6-leaf recursive partitioning Number of observed predict
decision tree generated using the above-mentioned 2D Class compounds % correct correct Enrichment
descriptors to classify the compounds. The structure of
the decision tree consists of six terminal and five non-ter- 1 12 60.00 46.15 66.67 1.08
minal nodes. At each decision point (node), molecules are 2 8 40.00 62.50 41.67 1.09
split into groups, higher and lower responses, according
to their descriptors. The terminal nodes 1, 3, and 5 are
class 1 (less active), while the terminal nodes 2, 4, and 6 terminal node 4 contains two compounds with density
are class 2 (active). The final tree model correctly classified <1.14 which is grouped in class 2 (active), while the
92 and 100% of the molecules in class 1 and 2 using remaining 6 compounds with density >1.14 are placed on
radius of gyration, principle moment of inertia, density, and the non- terminal node.
AlogP descriptors as decision points. A true response to
any given split follows the branch to the downside and a The fourth split appears on AlogP98, an atom-type-based
false response to any given split follows the branch to the logP method calculated using the latest published set of
upside in the decision tree. parameters (34). It is a thermodynamic descriptor that can
be used to relate chemical structure to observed lipophilic-
The first primary split is observed on radius of gyration and ity and its value is 3.26. It reflects the hydrophobic charac-
its value is 3.41. It is a spatial descriptor related to a mea- ter of the molecule. The terminal node 3 contains three
sure of the size of a molecule. The terminal node 6 con- compounds with AlogP98 <3.26 which are grouped in
tains five compounds with radius of gyration value <3.41 class 1 (less active), while the remaining three compounds
which is shown as class 2 (active), while the remaining 15 with AlogP98 >3.26 are placed on the 4th non-terminal
compounds have radius of gyration above 3.41 which node.
form the non-terminal node.
The last split occurs again on density and its value is 1.34.
The second split is observed on principal moment of iner- The terminal node 2 contains two compounds with density
tia, another spatial descriptor. In terminal node 5, seven value <1.34 which are categorized in class 2 (active), while
molecules have principal moment of inertia values <775.69 the remaining one compound has the density above 1.34
and are grouped in class 1 (less active). which forms the 1st terminal node.
The third split is observed on density, which is also a spa- To avoid over fitting and to improve generalization of the
tial descriptor and its value is 1.14. It reflects the type of classification models, internal validation was performed
atoms and how tightly they are packed in a molecule and using 10-fold cross-validation which resulted in 67% and
can also be related to transport and melt behavior. The 42% correct predictions for classes 1 and 2, respectively.
D.P, Sacchettini J.C. (2005) Synthesis, biological activ- study and recursive partitioning of heterocyclic quinone
ity, and X-ray crystal structural analysis of diaryl ether derivatives with antifungal activity. Bioorg Med
inhibitors of malarial enoyl acyl carrier protein reduc- Chem;14:1608–1617.
tase. Part 1: 4’-substituted triclosan derivatives. Bioorg 25. Young S.S., Hawkins D.M. (2004) Using recursive par-
Med Chem Lett;15:5247–5252. titioning analysis to evaluate compound selection
13. Mishra S., Karmodiya K., Parasuraman P., Surolia A., methods. Methods Mol Biol;275:317–334.
Surolia N. (2008) Design, synthesis, and application of 26. Rusinko A. 3rd, Young S.S., Drewry D.H., Gerritz S.W.
novel triclosan prodrugs as potential antimalarial and (2002) Optimization of focused chemical libraries using
antibacterial agents. Bioorg Med Chem;16:5536–5546. recursive partitioning. Comb Chem High Throughput
14. Perozzo R., Kuo M., Sidhu A.S., Valiyaveettil J.T., Bitt- Screen;5:125–133.
man R., Jacobs W.R. Jr, Fidock D.A., Sacchettini J.C. 27. Mamatha N., Murtuja S.B., Varaprasad B.F.M., Reddy
(2002) Structural elucidation of the specificity of the L.V., Bhattacharya A., Helliwell M., Mukherjee A.K.,
antibacterial agent triclosan for malarial enoyl acyl car- Beevi S.S., Mangamoori L.N., Mukkanti K., Pal S.
rier protein reductase. J Biol Chem;277:13106–13114. (2010) Naproxen and ibuprofen based acyl hydrazone
15. Khedkar V., Verma J., Coutinho E. (2009) Small-mole- derivatives: synthesis, structure analysis and cytotoxic-
cule screening techniques in drug discovery. Bio- ity studies. J Chem Pharm Res;2:393–409.
bytes;4:4–7. 28. Ershov A.Y., Lagoda I.V., Yakimovich S.I., Pakal’nis
16. Kitchen D.B., Decornez H., Furr J.R., Bajorath J. V.V., Zerova I.V., Dobrodumov A.V., Shamanin V.V.
(2004) Docking and scoring in virtual screening for (2009) Tautomerism and conformational isomerism of
drug discovery: methods and applications. Nat Rev mercaptoacetylhydrazones of aliphatic and aromatic
Drug Discov;3:935–949. aldehydes. Russ J Org Chem;45:660–666.
17. Friesner R.A., Banks J.L., Murphy R.B., Halgren T.A., 29. Palla G., Giovanni P., Domiano P. (1986) Conforma-
Klicic J.J., Mainz D.T., Repasky M.P., Knoll E.H., Shel- tional behaviour an E/Z isomerization of N-Acyl and N-
ley M., Perry J.K., Shaw D.E., Francis P., Shenkin P.S. aroylhydrazones. Tetrahedron;42:3649–3654.
(2004) Glide: a new approach for rapid, accurate 30. Hu €bschle C.B., Sheldrick G.M., Dittrich B.J. (2011)
docking and scoring. 1. Method and assessment of ShelXle: a Qt graphical user interface for SHELXL. J
docking accuracy. J Med Chem;47:1739–1749. Appl Crystallogr;44:1281–1284.
18. Halgren T.A., Murphy R.B., Friesner R.A., Beard H.S., 31. Tasdemir D., Lack G., Brun R., Ruedi P., Scapozza L.,
Frye L.L., Pollard W.T., Banks J.L. (2004) Glide: a new Perozzo R. (2006) Inhibition of Plasmodium falciparum
approach for rapid, accurate docking and scoring. 2. fatty acid biosynthesis: evaluation of FabG, FabZ, and
Enrichment factors in database screening. J Med FabI as drug targets for flavonoids. J Med
Chem;47:1750–1759. Chem;49:3345–3353.
19. Ramamurthy B., Bhatt M.V. (1989) Synthesis and antit- 32. Tasdemir D., Topaloglu B., Perozzo R., Brun R., O’Ne-
ubercular activity of N-(2-naphthyl)glycine hydrazide ill R., Carballeira N.M., Zhang X., Tonge P.J., Linden
analogues. J Med Chem;32:2421–2426. A., Ru €edi P. (2007) Marine natural products from the
20. Jain A.G., Gupta P., Ganesan G., Pande A., Malhotra Turkish sponge Agelas oroides that inhibit the enoyl re-
R. (2007) Rapid solvent-free synthesis of aromatic hy- ductases from Plasmodium falciparum, Mycobacterium
drazides under microwave irradiation. Def Sci J;57:267 tuberculosis and Escherichia coli. Bioorg Med
–270. Chem;15:6834–6845.
21. Awasthi S.R., Gupta P., Rao A., Ganesan K., Malhotra 33. Zmuidinavicius D., Japertas P., Petrauskas A., Didzia-
R. (2007) Synthesis, characterization and spectral petris R. (2003) Progress in toxinformatics: the chal-
studies of various newer long chain aliphatic acid (2- lenge of predicting acute toxicity. Curr Top Med
hydroxy benzylidene and 1H-indol-3-ylmethylene) hy- Chem;3:1301–1314.
drazides as mosquito para-pheromones. J Korean 34. Ghose A.K.V., Viswanadhan V.N., Wendolowski J.J.
Chem Soc;51:506–512. (1998) Prediction of hydrophobic (lipophilic) properties
22. Blower P., Fligner M., Verducci J., Bjoraker J. (2002) of small organic molecules using fragmental methods:
On combining recursive partitioning and simulated an analysis of ALOGP and CLOGP methods. J Phys
annealing to detect groups of biologically active com- Chem;102A:3762.
pounds. J Chem Inf Comput Sci;42:393–404.
23. Strobl C., Malley J., Tutz G. (2009) An introduction to Notes
recursive partitioning: rationale, application, and charac-
a
teristics of classification and regression trees, bagging, (1999–2009) iResarch Library. In: editor. iResarch Library.
and random forests. Psychol Methods;14:323–348. ChemNavigator.com. Inc.; p.
24. Choi S.Y., Shin J.H., Ryu C.K., Nam K.Y., No K.T., b
Accelrys I. (1998) Cerius2. In: editor. Cerius2. San Diego,
Park Choo H.Y. (2006) The development of 3D-QSAR CA, USA; p.