Discovery of potential 1,3,5-Triazine compounds against strains of Plasmodium falciparum using supervised machine learning models

Article abstract of DOI:10.1016/j.ejps.2019.105208

The Malaria burden was an escalating global encumbrance and need to be addressed with critical care. Anti-malarial drug discovery was integrated with supervised machine learning (ML) models to identify potent thiazolyl-traizine derivatives. This assimilated approach of Direct Kernel-based Partial Least Squares regression (DKPLS) with molprint 2D fingerprints in Quantitative Structure Activity Relationship models was utilized to map the knowledge of known actives and to design novel molecules. This QSAR study had revealed the structural features required for better antimalarial activity. Two of the molecules which were designed based on the results of this QSAR study, had shown good percentage of parasitemia against both chloroquine sensitive (3D7) and chloroquine resistant (Dd2) strains of Plasmodium falciparum respectively. The IC₅₀ of 201D and 204D was 3.02 and 2.17 μM against chloroquine resistant Dd2 strain of Plasmodium falciparum. This result had proved the efficiency of a multidisciplinary approach of medicinal chemistry and machine learning for the design of novel potent anti-malarial compounds.

Full text of DOI:10.1016/j.ejps.2019.105208

Journal Pre-proof
Discovery of Potential 1,3,5-Triazine Compounds Against Strains of
Plasmodium falciparum Using Supervised Machine Learning Models
Supriya Sahu , Surajit Kumar Ghosh , Jun Moni Kalita ,
Murali C Ginjupalli , Kranthi Raj K
PII:
S0928-0987(19)30481-6
DOI:
Reference:
https://doi.org/10.1016/j.ejps.2019.105208
PHASCI 105208
To appear in:
European Journal of Pharmaceutical Sciences
Received date:
Revised date:
Accepted date:
10 October 2019
29 November 2019
22 December 2019
Please cite this article as:
Supriya Sahu ,
Surajit Kumar Ghosh ,
Jun Moni Kalita ,
Murali C Ginjupalli , Kranthi Raj K , Discovery of Potential 1,3,5-Triazine Compounds Against
Strains of Plasmodium falciparum Using Supervised Machine Learning Models, European Journal of
Pharmaceutical Sciences (2019), doi: https://doi.org/10.1016/j.ejps.2019.105208
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.
©
2019 Published by Elsevier B.V.

Discovery of Potential 1,3,5-Triazine Compounds Against
Strains of Plasmodium falciparum Using Supervised Machine
Learning Models
a
a
a
b
c
Supriya Sahu* Surajit Kumar Ghosh Jun Moni Kalita Murali C Ginjupalli and Kranthi Raj K
a
Department of Pharmaceutical Sciences, Dibrugarh University, Dibrugarh 786004 Assam, India.
b
CaroCure Discovery Solutions Pvt. Ltd., 2897 Churchhill Lane Saginaw MI 48603, USA.
c
CaroCure Discovery Solutions Pvt. Ltd., IKP Knowledge Park, Genome Valley, Shamirpet, Hyderabad-500 101 Telangana
India.
*Corresponding author: supsjrt@gmail.com
Abstract
The Malaria burden was an escalating global encumbrance and need to be addressed with critical care.
Anti-malarial drug discovery was integrated with supervised machine learning (ML) models to identify
potent thiazolyl-traizine derivatives. This assimilated approach of Direct Kernel-based Partial Least
Squares regression (DKPLS) with molprint 2D fingerprints in Quantitative Structure Activity
Relationship models was utilized to map the knowledge of known actives and to design novel
molecules. This QSAR study had revealed the structural features required for better antimalarial
activity. Two of the molecules which were designed based on the results of this QSAR study, had
shown good percentage of parasitemia against both chloroquine sensitive (3D7) and chloroquine
resistant (Dd2) strains of Plasmodium falciparum respectively. The IC50 of 201D and 204D was 3.02
and 2.17 µM against chloroquine resistant Dd2 strain of Plasmodium falciparum. This result had
proved the efficiency of a multidisciplinary approach of medicinal chemistry and machine learning for
the design of novel potent anti-malarial compounds.
Keywords:
Anti-malarials
1
, 3, 5-triazine
Plasmodium falciparum
Machine Learning
QSAR

1
. Introduction
Based on the observation that Cycloguanil which was a potent antimalarial antifolate inhibits
Lactobacillus casei moderately [1, 2] the synthesis of 2-amino-4-phenyl thiazolyl-1, 3, 5- triazine
derivatives as antimalarial was started. Some of the thiazolyl-triazine derivatives synthesized by our
group for the first time showed encouraging inhibition of Lactobacillus casei [3-6]. Testing the
compounds against Lactobacillus casei clarifies the observation found between antibacterial and
antimalarial effect of Cycloguanil. These observations inspired us to design different thiazolyl-traizine
derivatives and to explore their antimalarial activities. Till now our group has synthesized hundreds of
such derivatives out of which, the thiazolyl-triazine core having secondary amine substitution on the
triazine ring and chlorine or nitro substitution on the second, third and/or fourth positions of the phenyl
thiazole ring (Table 1) exhibited better activity compared to the others [7, 8]. This finding is taken as
the backbone of the current study which is further analysed using machine learning (ML) models.
ML is the exercise of expending systems to analyze statistics, acquire knowledge from it and mark a
determination. It is also a forecast of the upcoming result of original data. Here the machine is taught
by means of statistics and algorithms that contribute it the capability to learn to accomplish the
assignment. Supervised and unsupervised learning are the two types of methods that can be used based
on the available data. Supervised learning approaches are used to progress training datasets to calculate
forthcoming values and these are all constant variables, while unsupervised procedures are used for
investigative drives to improve models that permit assembling of the statistics that are not indicated by
the user. So in this study we used supervised learning methods that train a model on known data that
can predict impending outputs using known inputs. The presence of in house generated data on
antimalarial activity prompted us to study 2D and 3D QSAR models for further refinement in their
structure and enhanced antimalarial activity.
Fifty seven thiazolyl triazine derivatives which has been reported by our group previously in various
journals [7-10] are been used for this QSAR study. Based on the results of the QSAR studies
favourable substitutions on 1, 3, 5-triazine ring was made with piperazine ring. The overlap of the 2D
and 3D visualization shows that this group has favourable effect on the activity. These analogues with
piperazine on triazine ring was synthesized and tested for the biological activity. 201D and 204D had
shown promising antimalarial activity against the Chloroquine resistant strain of Plasmodium
falciparum (Dd2).

2
2
. Material and methods
.1 2D & 3D QSAR
Data-driven or Machine learning methods are rapidly evolving and becoming very important in various
stages of drug discovery. The Supervised machine learning models are very significant in generating
the QSAR models. In this work we have applied supervised machine learning methods for the 2D and
3
D finger-prints of the anti-malarial compounds with respect to their biological activity to generate 2D
and 3D QSAR models respectively.
The first and significant step in these data driven method was the selection of the input samples
(
Training Set). Using supervised method, hierarchical clustering was used to cluster 57 antimalarial
compounds according to their chemical space (Table 1). Radial hashed fingerprint was used for the
clustering according to their chemical space. Compounds from each cluster with different ranges of
activity were selected as training set to cover chemical space and activity range. Thus 47 out of 57
compounds were selected as training set having activity (pIC50) range of 5.0 to 2.82 to build the 2D
QSAR model. Remaining 10 compounds with activity (pIC50) range of 4.92 to 2.82 were used as test
set to validate the hypothesis. There was a small change in the training and test sets of the 2D and 3D
QSAR models. This is because the 2 molecules alignment was not proper for the 3D QSAR and so
those were added to test set.
Table 1: List of compounds used in QSAR models
R1
NH
S
N
N
R2
N
N
N
N
R
H
H
Sl
No.
1
Compound
Code
A5
Structure
Activity Test/Training
2D
Test/Training
3D
R= 2, 4-dichloro; R
R= 3-nitro; R , R = -Phenyl
R= 2, 4-dichloro; R , R = -n-propyl
R= 3-nitro; R , R
R= 3-nitro; R , R
R= 2, 4-dichloro; R
R= 4-chloro; R =-H, R
R= 2, 4-dichloro; R , R
R= 2, 4-dichloro; R , R
1
, R
2
= -CH
3
4.999
4.945
4.904
4.874
4.818
4.736
4.65
Training
Training
Test
Training
Training
Test
2
H16
A3
1
2
3
1
2
4
H18
H2
1
2
= -piperidinyl
Training
Training
Training
Training
Test
Training
Training
Training
Training
Test
5
1
2
= -cyclopropyl
6
A29
66B
1
, R
2
= -benzyl
= piperazinyl
= -cyclopropyl
= -o-toluidininyl
7
1
2
8
A2
1
2
4.632
4.586
9
A21
1
2
Training
Training

1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
A20
79B
A19
68B
H5
R= 2, 4-dichloro; R
R= 4-chloro; R =aminoethyl, R
R= 2, 4-dichloro; R , R = -cyclohexyl
R= 4-chloro; R =hydroxyethyl, R = -H
R= 3-nitro; R , R = -CH
R= 3-nitro; R , R
R= 3-nitro; R , R
R= 3-nitro; R , R
R= 4-chloro; R
R= 2, 4-dichloro; R
R= -H; R , R = -Phenyl
R= 2, 4-dichloro ; R , R
R= 2, 4-dichloro; R , R
1
, R
2
= -p-toluidininyl
4.539
4.511
4.509
4.488
4.487
4.483
4.481
4.472
4.45
4.447
4.421
4.417
4.415
4.41
4.391
4.383
4.373
4.34
4.263
4.253
4.15
3.85
3.59
3.56
3.56
3.47
3.47
3.44
3.44
3.44
3.31
3.2
Training
Training
Training
Training
Training
Training
Test
Test
1
2
= -CH
3
Training
Training
Training
Test
1
2
1
2
1
2
3
H20
H19
H29
81B
A4
1
2
= -p-toluidininyl
= -cyclohexyl
= -benzyl
Training
Training
Training
Training
Training
Training
Training
Test
1
2
1
2
Training
Test
1
=-CH
3
, R
2
= piperazinyl
1
, R
2
= -dimethyl
Training
Training
Training
Training
Training
Training
Training
Test
E16
A16
A18
E3
1
2
1
2
= -Phenyl
1
2
= -piperidinyl
R= -H; R
R= -H; R
R= 3-nitro; R
R= -H; R , R
R= 3-nitro; R
R= 3-nitro; R
R= 4-chloro; R
R= 4-chloro; R
R= 4-chloro; R
R=3,4-dichloro; R
1
, R
2
= -n-propyl
Training
Training
Training
Test
E2
1
, R
2
= -cyclopropyl
H4
1
, R
2
= -dimethyl
E5
1
2
= -CH
3
H3
1
, R
2
= -n-propyl
Training
Test
Training
Test
H21
F5
1
, R
2
= -o-toluidininyl
1
, R
= -H, R
= -CH
= diethyl,R
2
= -CH
3
Training
Training
Training
Training
Training
Training
Training
Training
Training
Training
Training
Training
Training
Test
Training
Training
Training
Training
Test
67B
78B
34d
E18
E20
F21
84B
F2
1
2
= methylpiperazinyl
1
3
, R
2
= morpholinyl
= -SHC
1
2
4
H
9
R= -H; R
1
, R
2
= -piperidinyl
R= -H; R
1
, R
2
= -p-toluidininyl
Training
Training
Training
Training
Training
Training
Training
Test
R= 4-chloro; R
R= 4-chloro;R
R= 4-chloro; R
R= 4-chloro; R
R= 4-chloro; R
R= 4-chloro; R
R= -H; R , R = -o-toluidininyl
R= 4-chloro; R , R = -n-propyl
R=3,4-dichloro; R = diethyl, R
R= 4-chloro; R = diethyl, R = -n-butyl
R= 3, 4-dichloro; R = -H, R = -n-butyl
R= 4-chloro; R , R
R= 4-chloro; R , R
R= 4-chloro; R
1
, R
2
= -o-toluidininyl
1
= -CH
3
2
,R = hydroxyethylpiperazinyl
1
, R
2
= -cyclopropyl
F18
23c
82B
E21
F3
1
, R
2
= -piperidinyl
1
2
= diethyl, R = furan-2-methyl
1
= -CH , R = methylpiperazinyl
3
2
1
2
1
2
3.2
Training
Training
Test
28d
21c
12b
F4
1
2
= -Cl
3.2
Training
Test
1
2
3.13
3.11
3.1
1
2
Training
Training
Training
Test
Training
Training
Test
1
2
= -dimethyl
F19
72B
1
2
= -cyclohexyl
3.1
1
= -H, R
2
= -H
2.99
Training

4
5
5
5
5
5
5
5
5
9
0
1
2
3
4
5
6
7
4a
R= 4-chloro; R
R= 4-chloro; R
R=3,4-dichloro; R
R=3,4-dichloro; R
R= 4-chloro; R = -H, R
R= -H; R , R
R= -H; R , R
R= -H; R , R
R= 4-chloro; R
1
= -H, R
, R = -p-toluidininyl
= diethyl, R = isopropyl
= diethyl, R = -SHC
= hydroxyethylpiperazinyl
2
= -phenyl
2.98
2.98
2.94
2.92
2.82
2.82
2.82
2.82
2.82
Training
Training
Training
Training
Training
Training
Training
Training
Test
Training
Training
Training
Training
Training
Test
F20
29d
35d
69B
E4
1
2
1
2
1
2
6 5
H
1
2
1
2
= -dimethyl
= -cyclohexyl
= -benzyl
E19
E29
F29
1
2
Training
Training
Training
1
2
1
2
, R = -benzyl
Numerical representation of the structural features and atom typing schemes of input compounds
required for the quantitative prediction using QSAR techniques. The hashed fingerprints, Molprint2D
were used to generate the binary fingerprint for all the compounds used in this study. Molprint2D
considers each heavy atom in a structure and is characterized by an environment that consists of all
other heavy atoms within a distance of two bonds [11]. We have used supervised learning method that
couples Direct Kernel-based Partial Least Squares regression (DKPLS) [12] with molprint2D
fingerprints in Canvas software [13] to obtain predictive and interpretable 2D QSAR models.
Comparison with other eigenvector methods, DKPLS is highly efficient and its low rank approximation
tends to yield more robust regression models that can be visualized in terms of favourable and
unfavourable structural characteristics on the molecule. It produces the low rank approximation which
is suited well to regressions and in turn gives accurate results. Orthogonal factorization is used by
DKPLS regression to build low ranked estimates of kernel matrix K. These approximation and
estimates were used to evaluate regression in this work. Total emphasis was on how this technique is
adjusted to use through fingerprints generated using molprint along with QSAR visualization and
cunning of uncertainties in the estimates of the models. Gaussian kernel matrix with auto scaled
variable pairs denoted by column vectors X
ij = exp(-|| X
Atom based 3D QSAR models were developed based on our previous work on antimalarials [14].
i
& X
j
is defined as follows.
K
i
– X ||2 / 2σ2
j
2
.2 Chemistry
The QSAR study performed in this work was considered as the basis for the design of new potent
thiazolyl-triazine derivatives as antimalarial compounds. These compounds were synthesized using
scheme 1.

Scheme 1: Synthesis of thiazolyl-traizine derivatives
Cl
N
S
N
N
H N NH2
Thiourea
2
R
H N
2
N
Cl
Cl
R
O
Thionyl Chloride
S
0
Acetone, 0-5 C
60-90 mins
1
-(3,4-
2
-amino-4-substitued phenylthiazole
dichlorophenyl)ethanone
S
R
S
R
S
R
ammonia/
methylamine/
dimethylamine
HN
N
HN
N
HN
NR3
N
HN
N
N
N
N
N
N
5
min, 110ºC
µw,7PSI, Cl
R2
R2
20min, 40ºC
µw,3PSI
N
N
N
R1
N
N
R1
Cl
N
Cl
R N
1,4-dioxane
3
Compound code
01D
04D
R
R1
R2
R3
H
2
3, 4-Dichloro
3, 4-Dichloro
H
H
H
CH3
CH3
2
CH CH OH
2
2
H
H
CH3
7
A
1
44C
71C
4-CH3
4-CH3
CH3
CH2CH2OH
H
H
1
CH3
CH3
2
.3 Antimalarial activity evaluation
The chloroquine sensitive 3D7 and chloroquine resistant Dd2 strain of P. falciparum was routinely
maintained in medium, rose well park memorial institute-1640 (RPMI-1640) supplemented with 25
mmol 4-(2-hydroxyethyl)-1-piperazin-ethane sulphonic acid (HEPES), 1% D(+)-glucose, 0.23%
sodium bicarbonate and 10% heat inactivated human serum. The asynchronous parasites of P.
falciparum were synchronized after 5% D-sorbitol treatment to obtain only the ring stage parasitized
cells. The initial ring stage parasitaemia of 0.8–1.5% at 3% haematocrit in a total volume of 200 µL of
medium RPMI-1640 was uniformly maintained for carrying out the assay.
The in-vitro antimalarial assay was carried out with modified micro-assay of Reickmann and co-
workers [1] in 96 well-microtitre plates. A stock solution of 5 mg/mL of each of the test samples was
prepared in DMSO and subsequent dilutions were made with culture medium. The test compounds in
2
0 µL volume at 50 µg/mL concentration in duplicate well were incubated with parasitized cell
°
preparation at 37 C and 5% CO
2
level in a carbon dioxide incubator. After 36–40 h of incubation, the
blood smears were prepared from each well and stained with 3% Giemsa. The level of parasitemia in
terms of % dead rings along with schizonts was determined by counting a total of 100 asexual parasites
microscopically. Chloroquine was used as reference standard drug. Table 4 shows the in-vitro
antimalarial activity of the synthesized molecules.

3
3
3
. Results and Discussion
.1 QSAR modelling
.1.1 2D-QSAR modelling
The QSAR model was generated using 47 training set compounds using KPLS method with a
maximum of four latent factors based on standard deviation of regression (SD). The model was
validated using 10 test set compound for evaluating the predictive power of the model. The statistics of
the generated model was shown in the table 2. As listed in the table, at four factor KPLS model showed
2
a good internal correlation (R ) of 0.79 between the actual activity and predicted activity (figure 1) for
the training set. The model is having high predictive power for the external dataset having a correlation
2
(
Q ) of 0.78.
Table 2: 2D QSAR statistics for the 4 latent variables generated using KPLS method
Factors
SD
1
2
3
4
0.59
0.33
0.28
0.59
0.43
0.65
0.64
0.42
0.36
0.75
0.72
0.36
0.34
0.79
0.78
0.33
2
R
2
Q
RMSE
Figure 1: A) Actual and predicted activity of training set compounds obtained in 2D QSAR. B) Actual and predicted
activity of test set compounds obtained in 2D QSAR.
3
.1.2 2D QSAR Model Visualization
A visual analysis of atomic effects for Canvas KPLS models built from molprint 2D were shown in
Figure 2 for compounds with strong (H-bonded interaction) and weak affinities (ionic interaction) for
the antimalarial compounds in this study. Favourable atoms for the activity were coloured red, whereas
non-favourable atoms were coloured blue. Further the colour intensity reflects the strength of the effect.

In the highest active compound A5, the two amino methyl substitutions on the 1,3,5-triazine and the
second Carbon of di-chloro substituted phenyl were mapped with high intense red color indicating that
these groups were important for strong binding of antimalarial compounds with the target. Another
2
active compound H18, the anionic oxygen of NO group substituted on meta- position of phenyl ring
was showing strong red colour and contributed for higher activity. The nitrogen atom of the two
piperidine ring substituted on the 1,3,5-triazine ring was also contributing for the activity. The
hydrophobic carbon (blue color) of piperadine ring was not favouring the activity. This trend was also
observed in least active compounds F29 and 29C, the hydrophobic atoms on these compounds were
showing high intense blue colour and reduced activity.
Figure 2: 2D-QSAR visualization of highest active compounds A5 (A), H18 (B) and least active compounds 29d (C), F29
(D). Atoms mapped with red are favourable for activity and the blue are non-favourable for the activity.
3
.1.3 3D-QSAR Modelling
Further to explore the 3D features of the substitutions on the phenyl thiazol-2-yl-1,3,5-triazine-2 amine
scaffold and to quantify the activity, a supervised learning model using atom based properties a 3D
QSAR model was generated. During the generations of atom-based QSAR, all the compounds were
aligned according to similarity in their shape with respect to the highest active compound. Based on the
occupancy of type of atom of each aligned molecule in 3 dimensional grid space quantitative models
were generated. The types of atoms were hydrogen bond donors, hydrophobic/non-polar (H), negative

ionic (N), positive ionic (P) and electron-withdrawing (W) atoms. Each atom occupied in the grid space
was converted into bit string. The bit string was a collection of binary valued 3D descriptors and
considered as independent variables to generate a quantitative model using partial least squares (PLS).
For generating the models, 57 antimalarial compounds synthesized earlier by our group were divided
into 45 compounds as training set and 12 compounds as test set using unsupervised learning methods
(
described elsewhere in the paper). Using 5 latent variables and with leave-one-out (LOO) cross
validation method 3D QSAR model was generated and the predictive statistics was illustrated in the
table 3 and figure 3. As show in table 3, 5 factorial models were having the highest internal correlation
2
(
R ) of 0.82 and also external correlation of 0.81.
Table 3: Atom Based 3D-QSAR statistics for the 5 latent variables generated by using PLS method
Factors
SD
1
2
3
4
5
0.58
0.37
0.26
0.44
0.65
0.45
0.37
0.75
0.55
0.34
0.8
0.33
0.82
0.6
2
R
2
R CV
0.59
e-08
e-10
e-13
e-14
e-14
P
5.52
1.29
6.26
5.49
4.4
RMSE
0.66
0.1
0.4
0.31
0.8
0.33
0.78
0.95
0.3
2
Q
0.66
0.85
0.81
0.95
Pearson-r
0.34
0.93
Figure 3: The scatter plot between the actual activity (X-axis) vs predicted activity (Y-axis) obtained by using 5 latent
variables of atom based QSAR. A is the training set correlation and B is the test set correlation.
3
.1.4 3D-QSAR Visualisation

The 3D-countours generated by the atom based QSAR model showed the significance of the
substituents on the biological activity. Figure 4 illustrates the most significant favourable (blue) and
unfavourable interactions (red) when the two-factor QSAR model was applied to active (IC50 > 4.2 µM)
and inactive compounds (IC50 < 3.2 µM).
The large blue region around potential hydrogen bond donors on the substitutions of 1, 3, 5-triazine
ring suggest that these features were important for high activity. The compounds 66B, 79B were
showing higher activity due to the presence of donor group at this position. The blue contours of
electronegative, electro positive and electron withdrawing on the active compounds were overlapping
2
with the NO group of highly active compounds H16, H18 and H2. The small blue colour region near 2
and 4 substitution of 1, 3, 5-triazine ring suggested that smaller hydrophobic groups were tolerable at
this region and hence the N-methyl substitutions of the compound A5 and A3 were showing highest
activity. A large red region depicted the non-favourable interaction for activity. Most of the inactive
compounds were substituted with large hydrophobic groups at these positons and hence they were
weak binders. The blue colour path near ortho and meta-position of phenyl ring showed the favourable
region of the hydrophobic groups. The presence of the chlorine atom at this position tends to increase
the activity. But the non-favourable hydrophobic feature at para position of the phenyl ring decreases
the antimalarial activity for the compounds.

Figure 4: A- Hydrogen Bond Donor, B- electronegative, C- electron withdrawing, D- electro positive, E-, electro positive
and hydrophobic maps on active compounds. Blue colour contours indicates the favourable regions and red colour indicates
the non-favourable regions.
3
.1.5 Visualizations of 2D and 3D QSAR models on newly designed compounds
Based on the results of the 2D and 3D QSAR studies, favourable nitrogen atom was kept on 1, 3, 5-
triazine ring and the non-favourable hydrophobic atoms were substituted with favourable hydrogen
bond donors like substituted piperazine. We have tried piperazine bioisosteres like 2,6-diazaspiro [3.3]
heptane, (1S,4S)-2,5-diazabicyclo [2.2.1] heptane and octahydropyrrolo [3,4-c] pyrrole. Among all the
bioisosteres piperazine moiety has good synthetic feasibility, so preceded with piperazine ring. Further
as detected from 3D QSAR study, the favourable hydrophobic region at meta position of the phenyl
ring was obtained by substituting the hydrogen with chlorine atom to design new analogues. The
overlap of the 2D and 3D visualization (Figure 5) showed that this group has favourable effect on the
activity. Two analogues with piperazine on triazine ring and chlorine at meta position was synthesized
and tested for their in-vitro antimalarial activity. Further to validate our hypothesis we synthesized 3
more compounds which were predicting low activity with both the QSAR models and were tested.
Figure 5: Visualizations of 2D and 3D QSAR models on newly designed compounds; A. 2D visualization map; B are the
countor maps of 3D QSAR on the new design.
3
.2 Chemistry
N2-(4-(3,4-dichlorophenyl)thiazol-2-yl)-N4-methyl-6-(piperazin-1-yl)-1,3,5-triazine-2,4-diamine
201D): Physical state: yellow powder; % yield: 71; m.p.: 280ºC; R (silica gel G): 0.66 (ethyl
(
f

-
acetate:hexane, 1:1); Solubility: DMSO, ethanol, methanol, chloroform. UVmax (nm): 290.0; FTIR (cm
1
1
)
3369.20 (N-H sec, Str.); 3103.60 (C-H Str.); 1237.21, 1182.08 (CN, Ar.) H NMR (MeOD): δ ppm:
2
.5 (s, 3H, methyl), 2.9 (d, 4H, piperazine), 3.4 (d, 4H, piperazine), 6.7 (s, 1H, thiazole), 7.5 (d, 2H,
1
3
phenyl), 7.8 (d, 2H, phenyl). CNMR (MeOD): δ, ppm: 40.01, 122.38, 124.73, 126.67, 128.22, 129.40,
1
2
49.01, 152.21, 164.77, 167.32, 169.21. MS (EI) m/z 436.42 (M +1).
-(4-(4-(4-(3,4-dichlorophenyl)thiazol-2-ylamino)-6-(methylamino)-1,3,5-triazin-2-yl)piperazin-1-
yl)ethanol (204D): Physical state: yellow sticky solid; % yield: 65; m.p.: 218ºC; R
f
(silica gel G): 0.51
(
ethyl acetate:hexane, 1:1); Solubility: DMSO, ethanol, methanol, chloroform. UVmax (nm): 244.0;
-
1
FTIR (cm ) 3388.47 (O-H, str.); 3266.41 (N-H sec, Str.); 2988.98 (C-H Str.); 1238.79, 1187.04 (CN,
1
Ar.) H NMR (MeOH): δ ppm: 2.5 (s, 3H, methyl), 2.8 (d, 4H, piperazine), 3.7 (d, 4H, piperazine), 6.4
1
3
(
s, 1H, Thiazole), 7.5 (d, 2H, phenyl), 7.7 (d, 2H, phenyl). CNMR (MeOD): δ,ppm: 40.19,
2.77,125.54, 129.22, 131.90, 132.32, 152.11, 155.44, 165.44, 169.43, 171.32. MS (EI) m/z 479.82 (M
1).
5
+
6
-(4-methylpiperazin-1-yl)-N2-(4-phenylthiazol-2-yl)-1,3,5-triazin e-2,4-diamine (7A): Physical state:
o
brown sticky solid. Yield 62%, m.p.:40 C, R
f
value TLC (Hexane:Ethylacetate, 1:1) 0.83, Solubility:
-
1
Methanol, DMSO, ethanol, chloroform, UVmax (nm): 309.0; FTIR(cm ) 3449.34, 3359.67 (N-H
primary, Str.); 1640.02, 851.83(N-H primary, Bend.); 1451.92 (N-H secondary, Bend.); 2949.05
1
2
2
2
7
809.33 (C-H Str.); 1279.57, 1185.64 (CN, Ar.), H NMR (MeOD):δ ppm: 2.302 (m, 3H, methylene),
.41 (d, 4H, piperazine), 3.67 (d, 4H, piperazine), 6.66 (s, 1H, Thiazole), 7.25 (m, 1H, phenyl) 7.35 (d,
1
3
H, phenyl), 7.47 (d, 2H, phenyl). CNMR (MeOD): δ ppm: 42.99, 45.15-46.31, 55.21, 6.85, 76.87-
7.72, 101.62, 125.77-128.56, 134.67, 150.73, 168.27, MS (EI) m/z 369.12 (M +1).
2
-(4-(4-(methylamino)-6-(4-p-tolylthiazol-2-ylamino)-1,3,5-triazin-2-yl)piperazin-1-yl)ethanol (144C):
f
Physical state: yellow sticky; % yield: 79; m.p.: 45ºC; R (silica gel G): 0.74 (ethyl acetate:hexane,
-
1
1
1
:1); Solubility: DMSO, ethanol, methanol; UVmax (nm): 233.0; FTIR (cm ) 3301.66 (O-H, Str.);
1
592.84 (N-H sec, Bend.); 2941.19, 2820.99 (C-H Str.); 1303.26, 1262.86 (CN, Ar.) H NMR (MeOD):
δ ppm: 2.2 (s, 3H, methyl), 2.4 (s, 3H, methyl), 2.7 (d, 4H, piperazine), 3.9 (d, 4H, piperazine), 6.6 (s,
1
3
1
1
H, Thiazole), 7.4 (d, 2H, phenyl), 7.6 (d, 2H, phenyl). CNMR (MeOD): δ,ppm: 40.22, 52.09,
21.22, 123.22, 125.01, 126.59, 127.71, 147.67, 150.98, 163.02, 166.65, 167.88. MS (EI) m/z 426.21
(
M +1).

N2,N2-dimethyl-6-(piperazin-1-yl)-N4-(4-p-tolylthiazol-2-yl)-1,3,5-triazine-2,4-diamine
(171C):
Physical state: yellow sticky; % yield: 73; m.p.: 68ºC; R (silica gel G): 0.65 (ethyl acetate:hexane,
f
-
1
1
1
:1); Solubility: DMSO, ethanol, methanol; UVmax (nm): 277.0; FTIR (cm ) 3361.61 (N-H sec, Str.);
1
560.21 (N-H sec, Bend.); 2922.17, 2865.72 (C-H Str.); 1367.96, 1230.58 (CN, Ar.) H NMR (MeOD):
δ ppm: 2.2 (s, 3H, methyl), 2.5 (s, 3H, methyl), 3.2 (d, 4H, piperazine), 3.7 (d, 4H, piperazine), 6.7 (s,
1
3
1
1
H, Thiazole), 7.6 (d, 2H, phenyl), 7.8 (d, 2H, phenyl). CNMR (MeOD): δ,ppm: 40.29, 51.89,
24.77, 128.33, 129.45,130.98, 150.62, 164.79, 167.19, 168.21. MS (EI) m/z 396.37 (M +1).
3
.3 Antimalarial activity evaluation
Table 4: In -vitro antimalarial activity of the synthesized molecules
Molecule
% parasitemia
3D7)
µg/ml 50 µg/ml 5 µg/ml 50 µg/ml
% parasitemia
(Dd2/RKL2)
IC50
(Dd2/RKL2)
(µM)
pIC50
Predicted Predicted
Activity – Activity –
2D QSAR 3D QSAR
(
5
2
2
01D
04D
98.5
96.0
21.5
31
100
98.5
46.5
47
71.0
100
3.02
5.52
5.66
4.35
4.41
4.40
4.92
4.95
3.94
4.17
3.89
5.14
5.23
4.75
4.03
4.14
69.0
100
2.17
*
*
*
7A
17.5
32.5
16.5
13.5
43.8
*
*
*
*
1
1
44C
71C
0
38.43
*
*
19
47
0
39.24
*
The strain used was RKL2 (One of the resistant strains of P. falciparum)
Molecules that were designed based on the observations from 2D & 3D QSAR had shown good anti-
malarial activity. All the molecules were tested against sensitive and resistant strains of P. falciparum.
Among five molecules 201D and 204D had shown 100% inhibition on resistant strain (Dd2) of P.
falciparum. Both the molecules were highly active on same resistant strain (Dd2) with IC50 of 3.02 and
2
.17 µM. As predicted remaining three molecules (7A, 144C and 171C) were depreciated in biological
activity when compared to active molecules. In-vitro activities of all the synthesized molecules were
shown in table 4.
4
. Conclusion
The 2D and 3D QSAR models generated by the machine learning approach to understand the relevance
of different substitutions on the thiazolyl-triazine core has suggested the presence of favourable
secondary amino nitrogen and replacement of the non-favourable hydrophobic atoms with favourable
hydrogen bond donors like substituted piperazine on triazine ring. Also, the favourable hydrophobic
interaction of the phenyl ring was obtained by substituting the hydrogen with chlorine atoms at third
and fourth positions. These findings of the ML approach was further validated by the synthesis of five

compounds, two having chlorine substitution on the phenyl ring whereas the other three was having
hydrogen or methyl groups attached to the phenyl ring. The two thiazolyl-traizine derivatives, 201D
and 204D proved to be active against resistant strains of Plasmodium falciparum and can be further
developed as next level anti-malarials to diminish the global burden using the model detailed in this
study. This interdisciplinary process of combining medicinal chemistry, supervised machine learning
and QSAR had steered the advancement of alternate approaches in antimalarial drug discovery process.
Acknowledgements
The financial support from the Department of Biotechnology (DBT), New Delhi, India is gratefully
acknowledged [grant no. BCIL/NER-BPMC/2013-541 (BT/330/NE/TBP/2012 dated 29 April 2013)].
The authors also are thankful to S.A.I.F., Punjab University, Chandigarh, India for providing
spectroscopic data; RMRC-Lahoal, Dibrugarh for providing the facilities for carrying out the
antimalarial assays; Dibrugarh University for providing the infrastructure and also to Schrodinger for
providing the software.
References
[
1] Rieckmann K.H., Sax L.J., Campbell G.H., Mrema J.E., 1978. Drug sensitivity of P. falciparum: an
in vitro microtechnique. Lancet. 1, 22–23.
2] Genther C.S., Smith C.C., 1977. Antifolate studies. Activities of 40 potential antimalarial
[
compounds against sensitive and chlorguanide triazine resistant strains of folate-requiring bacteria
and Escherichia coli. J. Med. Chem. 20, 237–243.
[
3] Gahtori P., Singh A., Ghosh S.K., Das A., Archana U., 2011. Synthesis of some
substituted phenylthiazolyl 1, 3, 5-triazine derivatives. Asian J. Chem. 23, 1189–1192.
4] Gahtori P., Ghosh S.K., Singh B., Singh U.P., Bhat H.R., Uppal A., 2011. Synthesis, SAR and
antibacterial activity of hybrid chloro, dichloro-phenylthiazolyl-s-triazines. Saudi Pharm. Journ. 20,
5–43.
5] Gahtori P., Ghosh S.K., 2011. Design, synthesis and SAR exploration of hybrid 4-
chlorophenylthiazolyl-s-triazine as potential antimicrobial agents. J. Enz. Inhib. and Med. Chem. 1–13.
6] Gahtori P., Ghosh S.K., Parida P., Prakash A., Gogoi K., Bhat H.R., Singh U. P., 2012. Antimalarial
[
3
[
[
evaluation and docking studies of hybrid phenylthiazolyl-1,3,5-triazine derivatives: A novel and
potential antifolate lead for Pf-DHFR-TS inhibition. Ex. Para. 130, 292–299.

[
7] Kalita J.M., Ghosh S.K., Sahu S., Dutta M., 2017. Rational design and microwave assisted
synthesis of some novel phenyl thiazolyl clubbed s-triazine derivatives as antimalarial antifolate. Fut. J.
Pharm. Scs. 3, 11-17.
[
8] Sahu S., Ghosh S.K., Gahtori P., Singh U.P., Bhattacharyya D.R., Bhat H.R., 2019. In silico
ADMET study, docking, synthesis and antimalarial evaluation of thiazole-1,3,5-triazine derivatives as
Pf-DHFR inhibitor. Pharmacolog. Report. 71, 762–767.
[
9] Sahu S., Ghosh S.K., Ghoshal A., Kalita J., Gahtori P., Bhattacharyya D.R., 2016. Microwave
assisted synthesis, antimalarial screening and structure–activity-relationship exploration of some
phenylthiazolyl-triazine derivatives against dihydrofolate reductase. Med. Chem. Res. 25, 2916-2923.
[
10] Gahtori P., Pandey R., Kumar V., Ghosh S.K., Das A., Kalita J.M., Sahu S., Prakash A.,
Bhattacharyya D.R., 2016. Toward resistance-compromised DHFR inhibitors part 1: combined
structure/ ligand-based virtual screenings and ADME-Tox profiling. J. Chemomet. 30, 462-481.
[
11] Bender A., Mussa H.Y., Glen R.C., Reiling S., 2004. Similarity searching of chemical databases
using atom environment descriptors (MOLPRINT 2D): evaluation of performance. J. Chem. Inf.
Comput. 44, 1708–1718.
[
12] An Y., Woody S., Steven D.L., 2013. Kernel-Based Partial Least Squares: Application to
Fingerprint-Based QSAR with Model Visualization. J. Chem. Inf. and Mod. 53, 2312-2321.
[
[
13] Canvas, version 1.5; Schrodinger L.L.C.: New York, 2012.
14] Raj K.K., Manohar S., Talluri V.R., Rawat D.S., 2015. Insights into activity enhancement of 4-
aminoquinoline-based hybrids using atom-based and field-based QSAR studies. Med. Chem. Res. 24,
136-1154.
1

Graphical Abstract