Professional Documents
Culture Documents
A R T I C L E I N F O A B S T R A C T
Keywords: Obesity has acquired notable attention due to its high occurrence and link with grievous health problems such as
QSAR hypertension, diabetes and heart disease. It has been reported that the endocannabinoid system executes a pivotal
SMILES part in the management of food absorption, fat augmentation, and energy balance. In the present manuscript, we
CORAL
report a detailed QSAR analysis for 165 CB1 cannabinoid receptor inhibitors employing the Monte Carlo opti-
CB1 receptor inhibitors
IIC
mization process incorporated within the CORAL software. Eight splits are made from the whole dataset and
CCI sixteen QSAR models are developed from these splits employing two target function TF1 (without index of ideality
of correlation) and TF2 (with index of ideality of correlation). All the QSAR models developed with TF2 have
better predictive potential than the models developed with TF1. The model built for split 5 using TF2 is the leading
model due to the higher value of the determination coefficient of the validation set (R2Valid ¼ 0.8518). The index of
ideality of correlation (IIC) improves the statistical performance of CORAL-based QSAR-models and gives sta-
tistically robust predictive models of the investigated endpoint pIC50. In the present manuscript, a novel criterion
“Correlation Contradiction Index (CCI)” is also applied to know its predictive potential. The absolute value of CCI
for calibration set is less when QSAR models are developed employing IIC. The promoters of increase and decrease
endpoint pIC50 are identified and these are applied to design seven new compounds. All the newly designed
molecule were docked into in the active site of human cannabinoid receptor CB1 (PDB ID: 5tgz).
* Corresponding author.
E-mail addresses: parvinchem@kuk.ac.in, parvinjangra@gmail.com (P. Kumar).
https://doi.org/10.1016/j.chemolab.2020.103982
Received 23 July 2019; Received in revised form 24 February 2020; Accepted 28 February 2020
Available online 5 March 2020
0169-7439/© 2020 Elsevier B.V. All rights reserved.
P. Kumar, A. Kumar Chemometrics and Intelligent Laboratory Systems 200 (2020) 103982
Table 1
Percentage of the identity of splits 1–8 CB1 receptor inhibitors.
Split SET Split 1 (%) Split 2 (%) Split 3 (%) Split 4 (%) Split 5 (%) Split 6 (%) Split 7 (%) Split 8 (%)
Split 1 Total 100 0.0 0.0 0.0 25.5 24.8 24.2 25.5
Training 100 0.0 0.0 0.0 24.4 26.5 24.4 24.4
Invisible training 100 0.0 0.0 0.0 26.2 24.1 24.1 26.5
Calibration 100 0.0 0.0 0.0 26.5 24.1 23.8 26.5
Validation 100 0.0 0.0 0.0 24.7 24.7 24.7 24.4
2
P. Kumar, A. Kumar Chemometrics and Intelligent Laboratory Systems 200 (2020) 103982
further divided into four sets i.e. training set, invisible training set, The correlation weights in term of numerical data which give better
calibration set and validation set. Each set has its individual account- statistics for calibration set is calculated from the training and invisible
ability in QSAR formation. The reported method [23,45]was applied to training set is represented by the following equation:
know the non-identical character of splits and it is confirmed by Table 1. SMILES
The graphical presentation of split distribution is given in Fig. 1. Each set Endpoint ¼ C0 þ C1 DCW (T, Nepoch) (2)
has a discrete assignment to execute [25,46]. The assignment of the The predictive potential of the model is verified by the external
training set was to make the QSAR model by calculating the correlation validation set.
weights and the assignment of the invisible training set was to test the In the CORAL software 2019, three types of target functions i.e. (i) the
fitness of the compounds which were not present in training set. The classic scheme; (ii) balance of correlation; (iii) Index of Ideality of Cor-
work assigned to the calibration set was to identify the start of over- relation (IIC) are designed. IIC has been recommended as a novel
training whereas the job of the validation set was to verify the models for benchmark and it is applied to judge the predictive potential of devel-
the compounds which were not present in all the three sets i.e. training oped QSAR models. In this research paper, we use two target functions
set, invisible training set and calibration set [47,48]. for Monte Carlo optimization: balance of correlation (TF1) and balance of
Identity (%)¼ [Nij/0.5(NiþNj)]x100 correlation with IIC (TF2)
X X X
DCW T; Nepoch ¼ CWðSK Þþ CWðSSK Þ þ CWðSSSK Þ þ CWðBONDÞ þ CWðNOSPÞþ
(1)
CWðHALOÞ þ CWðHARDÞ þ CWðPAIRÞ þ CWðCmax Þ þ CWðNmax Þ þ CWðOmax Þ þ CWðSmax Þ
Here, CW(Z) is the correlation weight for a SMILE descriptor/attri- MAE is mean absolute error which can be determined with the
butes. T and Nepoch stand for threshold and number of epoch respectively. following equation:
Threshold coefficient is applied to classify various molecular attributes
1 X
N
obtained from SMILES into two classes: (i) active class i.e. CW is used in
M AEset ¼ jΔk j Δk < 0;N is the number of Δk < 0 (6)
the model making and (ii) rare class i.e. CW is not used in the model N k¼1
making. The best statistical quality for the calibration set is given by the
Nepoch. The detailed description of global SMILES attributes is given in
Table 2 [49].
3
P. Kumar, A. Kumar Chemometrics and Intelligent Laboratory Systems 200 (2020) 103982
Table 2 Table 3
The detailed description of SMILES attributes. The mathematical equation of various statistical parameters.
S.No. SMILES Comments S. No. Equation
notation P
1 ðYobs Ypred Þ2
1. Sk SMILES-atoms, i.e. one symbol or two symbols which R2 ¼ 1 P
ðYobs YÞ2
cannot be examined separately P
2 ðYobs Ypred Þ2
2. SSk A combination of two SMILES-atoms 2
QLOO ¼ 1 P
ðYobs YÞ2
3. SSSk A combination of three SMILES-atoms 8 92
3 > P >
4. BOND The presence or absence of double (‘ ¼ ’), triple (‘#’) and < next =
i¼1 ðYobs Y obs ÞðYpred Y pred Þ
stereochemical (‘@’) bonds R2ext ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P ffi
>
: next 2 2> ;
5. NOSP Presence or absence of nitrogen, oxygen, sulfur and i¼1 ðY obs Y obs Þ ðY pred Y pred Þ
phosphorus Pnext 2
4 ðY obs Y pred Þ
6. HALO Presence or absence of halogens
2
QF1 ¼ 1 Pni¼1 2
i¼1 ðYobs Y train Þ
ext
9. Nmax Total number of nitrogen atoms in the molecular structure 6 minðMAEset; MAEsetÞ
IIC ¼ Rset
10. Omax Total number of oxygen atoms in the molecular structure maxðMAEset; MAEsetÞ
11. Smax Total number of sulfur atoms in the molecular structure 7 Pntest ðyi b y i Þ2
i¼1
ntest RMSEP2
2
QF3 ¼ 1 ¼ 1
Pntest ðyj b y TR Þ2 S2TR
1 X
N j¼1
nTR
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
M AEset ¼ jΔk j Δk < 0; þ N is the number of Δk < 0: (7) 8 2
rmðtestÞ ¼ r 2 ð1 r 2 r 2o Þ
N k¼1 P
9 CCIcal ¼ ΔRj cal ; ΔRj cal < 0
P ext
The quality of prediction (Δk) for one substance from a set can be 10 2 ni¼1 ðYobs Y obs ÞðYpred Y pred Þ
CCC ¼ Pnext 2 Pnext 2 2
estimated as the following: i¼1 ðYobs Y obs Þ þ i¼1 ðYpred Y pred Þ þ next ðYobs Y pred Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
11
CR2p ¼ R ðR Rr Þ
2 2
Δk ¼ observedk – calculatedk (8)
12 1 P
MAE ¼ Yobs Ypred
n rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
13 1 X
RMSEP ¼ ðYobs Ypred Þ2
n
2.3. Correlation Contradiction Index (CCI)
the validation of the generated QSAR models [53–61]. The mathematical
The Correlation Contradiction Index (CCI) is a new criterion and it equation of various statistical parameters is given in Table 3. The IIC was
works on the hypothesis “The smaller absolute value of CCI, the better also implemented to validate the predictive potential and to judge the
predictive potential of a model (i.e. statistical characteristics for the better model for a developed QSAR model [19,29,50,51,62]. Finally, the
validation set should be better if absolute value of CCI is small)” given by novel criterion, CCI was also used to validate the built QSAR models.
Toropova et al. [52]. The CCI is the sum of all negative contributions over
the calibration set and it is represented by the following mathematical
equation 2.5. Applicability domain
X
CCIcal ¼ ΔRj cal ; ΔRj cal < 0 (9) From the list of various standard statistical parameter, applicability
domain (AD) is one of the most significant parameters which is procured
ΔRj cal ¼ R2cal R2j cal (10) from the training set data of a particular in silico model. The AD is
summarized in the groups of structural, biological, physicochemical
where R2cal ¼ determination coefficient for all objects for calibration set; space, data or information and mechanism of action for which the model
of the training set is made and it can be applied for the evaluation of the
R2j cal ¼ correlation coefficient calculated without the jth object.
reliability of a built QSAR model. AD can be applied to judge the pre-
The jth object is “supporter” if the ΔRj cal is positive and its aid to the dictability of a developed QSAR model for the molecules which are not
correlation coefficient of the calibration set. Vice versa, the jth object is considered in making of the model. AD is useful to recognize the inter-
“oppositionist” if the ΔRj cal is negative and it decreases the correlation polation (true predictions) or extrapolation (less reliable predictions) of
coefficient of the calibration set. the developed QSAR model. However, in the case of the Monte Carlo
method based QSAR by CORAL, the statistical defects of SMILES are the
2.4. Validation principles according to which AD is defined. The “statistical defect,” d(A)
for an attribute of SMILES is helpful to understand the robustness of the
The main target of the developed QSAR model is to build a robust model and it can be represented by the following mathematical equation:
model which is efficient to predict the properties of novel molecule in an
jPðAÞ P’ðAÞj
objective, authentic and precise manner. The judgment of robustness and dðAÞ ¼ (11)
NðAÞ þ N’ðAÞ
reliability of a built QSAR model can be made on the basis of the
following approach: a) internal validation or cross-validation taking into Here, P(A) & P0 (A) are the probabilities of an attribute (A) in the
account the training set data, b) external validation taking into account training and calibration set respectively; N(A) and N0 (A) are the number
the test set data and c) data randomization or Y-scrambling. The Monte of times attribute (A) appears the above sets.
Carlo based QSAR models are successfully validated by applying this For a given molecule, the statistical defect (D) is the summation of the
approach. statistical defect, d(A), of all the features available in the SMILES
Various standard statistical parameters such as correlation coefficient notation
(r2), concordance correlation coefficient (CCC), cross-validated correla-
tion coefficient (q2), Q2F1, Q2F2, Q2F3, standard error of estimation (s), X
NA
D ¼ defectðSMILESÞ ¼ dðAÞ (12)
mean absolute error (MAE), Fischer ratio (F) and root-mean-square error k¼1
(RMSE), and novel metrics (Rm2 and MAE based metric) are applied for
4
P. Kumar, A. Kumar
Table 4
The various statistical parameter of built QSAR models with their comparisons for eight splits using TF1 and TF2.
Split Target SET n R2 CCC IIC Q2 Q2F1 Q2F2 Q2 F 3 s MAE F F (0.05, 1, p Y-test CR2p Avg Rm2 ΔRm2 Equation
Function n-2)
1 TF1 Training 41 0.9325 0.9651 0.8340 0.9249 0.228 0.184 539 251.06 0.034 0.0245 0.9202 pIC50 ¼ 3.9770599 (0.0310344) þ
InvTrain 42 0.9449 0.9697 0.8245 0.9385 0.231 0.176 686 251.14 0.030 0.0199 0.9349 0.0983955 (0.0007412) *
Calib 42 0.7136 0.8392 0.7846 0.6734 0.6408 0.6403 0.6217 0.548 0.413 100 251.14 0.079 0.0335 0.6966 0.5993 0.1573 DCW(10,6)
Valid 40 0.8033 0.8962 0.8447 0.7782 0.392 0.319 155 250.98 0.064 0.0307 0.7878 0.7205 0.0067
TF2 Training 41 0.8421 0.9143 0.8740 0.8230 0.348 0.267 208 251.06 0.055 0.0169 0.8336 pIC50 ¼ 2.4315214 (0.0748214) þ
InvTrain 42 0.9218 0.9538 0.7685 0.9128 0.266 0.218 471 251.14 0.036 0.0284 0.9075 0.0858929 (0.0011069) *
Calib 42 0.8045 0.8799 0.8969 0.7843 0.7975 0.7972 0.7868 0.412 0.318 165 251.14 0.062 0.0256 0.7916 0.6574 0.1914 DCW(3,12)
Valid 40 0.8187 0.8936 0.5760 0.7966 0.378 0.300 172 250.98 0.060 0.0303 0.8034 0.7376 0.1532
Comments Correct Incorrect Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Incorrect
2 TF1 Training 42 0.9674 0.9834 0.8941 0.9643 0.164 0.112 1186 251.14 0.023 0.0167 0.9590 pIC50 ¼ 1.9308274 (0.0274645) þ
InvTrain 41 0.9663 0.9795 0.9805 0.9618 0.171 0.125 1117 251.06 0.024 0.0411 0.9455 0.1330787 (0.0005774) * DCW(1,3)
Calib 40 0.7784 0.8710 0.7372 0.7557 0.7748 0.7748 0.7915 0.407 0.344 134 250.98 0.068 0.0377 0.7593 0.6673 0.1980
Valid 42 0.7126 0.8420 0.8310 0.6794 0.497 0.364 99 251.14 0.079 0.0215 0.7018 0.6004 0.1020
TF2 Training 42 0.9316 0.9646 0.9652 0.9246 0.237 0.169 545 251.14 0.034 0.0170 0.9231 pIC50 ¼ 1.2023358 (0.0499522) þ
InvTrain 41 0.8907 0.9385 0.9029 0.8794 0.296 0.237 318 251.06 0.044 0.0145 0.8834 0.1078466 (0.0007547) *
Calib 40 0.8917 0.9406 0.9436 0.8779 0.8898 0.8898 0.8979 0.285 0.219 313 250.98 0 .045 0.0170 0.8832 0.8259 0.1036 DCW(2,13)
Valid 42 0.7368 0.8524 0.7964 0.7104 0.470 0.382 112 251.14 0.075 0.0233 0.7251 0.6283 0.1692
Comments Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct
3 TF1 Training 40 0.8603 0.9249 0.9275 0.8436 0.321 0.254 234 250.98 0.052 0.0236 0.8484 pIC50 ¼ 3.8816131 (0.0509205) þ
InvTrain 42 0.8778 0.9282 0.7006 0.8600 0.328 0.258 287 251.14 0.047 0.0354 0.8599 0.1307608 (0.0015534) * DCW(6,2)
Calib 42 0.8359 0.9059 0.7129 0.8166 0.8342 0.8342 0.8272 0.369 0.281 204 251.14 0.055 0.0195 0.8261 0.7233 0.1575
Valid 41 0.7284 0.8527 0.7843 0.6992 0.467 0.364 105 251.06 0.077 0.0190 0.7188 0.6217 0.0574
5
TF2 Training 40 0.8794 0.9359 0.8485 0.8655 0.298 0.229 277 250.98 0.048 0.0189 0.8699 pIC50 ¼ 2.8318831 (0.0568819) þ
InvTrain 42 0.8682 0.9225 0.7112 0.8504 0.337 0.277 264 251.14 0.049 0.0197 0.8583 0.1401427 (0.0014856) * DCW(4,6)
Calib 42 0.8188 0.9016 0.9035 0.7972 0.8132 0.8132 0.8053 0.392 0.296 181 251.14 0.059 0.0076 0.8150 0.7405 0.0790
Valid 41 0.7379 0.8571 0.7858 0.7084 0.470 0.353 110 251.06 0.075 0.0178 0.7289 0.6345 0.0058
Comments Correct Correct Correct Incorrect Incorrect Incorrect Incorrect Incorrect Incorrect Incorrect Incorrect Correct Incorrect correct correct
4 TF1 Training 42 0.7961 0.8865 0.8922 0.7665 0.413 0.326 156 251.14 0.063 0.0193 0.7864 pIC50 ¼ 3.6452627 (0.0785066) þ
InvTrain 40 0.9038 0.9364 0.5578 0.8940 0.298 0.218 357 250.98 0.042 0.0094 0.8991 0.0919756 (0.0016254) * DCW(8,3)
TF2 Training 42 0.9126 0.9543 0.7892 0.8994 0.270 0.208 418 251.14 0.038 0.016 0.9044 pIC50 ¼ 1.0874582 (0.0761792) þ
InvTrain 40 0.9273 0.9454 0.3995 0.9193 0.298 0.231 485 250.98 0.036 0.017 0.9187 0.1013066 (0.0011163) *
Calib 41 0.7401 0.8570 0.8600 0.7139 0.6888 0.6884 0.6960 0.490 0.378 111 251.06 0.075 0.022 0.7289 0.6347 0.1257 DCW(2,14)
Valid 42 0.8220 0.9035 0.6921 0.7971 0.394 0.310 185 251.14 0.058 0.026 0.8088 0.7456 0.0275
Comments Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Incorrect Correct Correct Incorrect
5 TF1 Training 41 0.8821 0.9374 0.8945 0.8688 0.287 0.209 292 251.06 0.046 0.0399 0.8620 pIC50 ¼ 1.6195651 (0.0652232) þ
InvTrain 42 0.8802 0.9365 0.7799 0.8692 0.287 0.233 294 251.14 0.046 0.0427 0.8586 0.1138487 (0.0011623) * DCW(9,2)
Calib 41 0.7841 0.8776 0.6582 0.7587 0.7726 0.7588 0.6793 0.462 0.342 142 251.06 0 .066 0.0155 0.7764 0.6944 0.0605
Valid 41 0.8311 0.9094 0.7039 0.8081 0.389 0.315 192 251.06 0.057 0.0223 0.8199 0.7567 0.0967
TF2 Training 41 0.8767 0.9343 0.8918 0.8638 0.294 0.226 277 251.06 0.047 0.0209 0.8662 pIC50 ¼ 1.4915892 (0.0633723) þ
InvTrain 42 0.8720 0.9319 0.8522 0.8576 0.285 0.225 273 251.14 0.048 0.0123 0.8659 0.0755870 (0.0007347) * DCW(3,9)
Calib 41 0.8413 0.9010 0.9172 0.8237 0.8427 0.8332 0.7783 0.384 0.267 207 251.06 0.055 0.0285 0.8269 0.6813 0.1685
Valid 41 0.8518 0.9107 0.8864 0.8331 0.374 0.327 224 251.06 0 .053 0.0232 0.8401 0.7440 0.1440
Comments Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Incorrect Correct Incorrect Incorrect
6 TF1 Training 42 0.9584 0.9788 0.8088 0.9548 0.162 0.119 922 251.14 0.026 0.0357 0.9404
(continued on next page)
P. Kumar, A. Kumar
Table 4 (continued )
Split Target SET n R2 CCC IIC Q2 Q2F1 Q2F2 Q2 F 3 s MAE F F (0.05, 1, p Y-test CR2p Avg Rm2 ΔRm2 Equation
Function n-2)
InvTrain 41 0.9331 0.9432 0.8925 0.9246 0.308 0.260 544 251.06 0.033 0.0192 0.9235 pIC50 ¼ 0.4873158 (0.0377398) þ
Calib 41 0.8389 0.9158 0.7818 0.8226 0.8474 0.8338 0.7786 0.384 0.301 203 251.06 0.056 0.0292 0.8242 0.7688 0.0208 0.1521567 (0.0006791) *
Valid 41 0.6395 0.7901 0.6863 0.5629 0.654 0.429 69 251.06 0.095 0.0317 0.6234 0.5062 0.1784 DCW(4,10)
TF2 Training 42 0.8380 0.9118 0.8322 0.8193 0.320 0.249 207 251.14 0.055 0.0477 0.8138 pIC50 ¼ 2.7870194 (0.0641000) þ
InvTrain 41 0.8147 0.9024 0.7803 0.7933 0.367 0.261 172 251.06 0.060 0.0561 0.7862 0.1057928 (0.0013248) * DCW(3,6)
Calib 41 0.8634 0.9109 0.9284 0.8465 0.8538 0.8408 0.7879 0.376 0.316 246 251.06 0.050 0.0271 0.8497 0.7639 0.1329
Valid 41 0.7136 0.8374 0.7536 0.6831 0.509 0.350 97 251.06 0.080 0.0270 0.7000 0.5994 0.1533
Comments Correct Correct Correct Correct Correct Correct Correct Correct Incorrect Correct Correct Correct Correct Incorrect Incorrect
7 TF1 Training 41 0.9059 0.9506 0.8220 0.8950 0.289 0.242 375 251.06 0.040 0.0118 0.8999 pIC50 ¼ 0.2012223 (0.0726707) þ
InvTrain 41 0.9124 0.9486 0.8030 0.9005 0.286 0.214 406 251.06 0.039 0.0183 0.9032 0.0965379 (0.0008883) *
Calib 42 0.6716 0.8117 0.4452 0.6372 0.6750 0.6381 0.7419 0.478 0.335 82 251.14 0.087 0.0107 0.6663 0.5497 0.0829 DCW(10,3)
Valid 41 0.6639 0.8045 0.5928 0.6283 0.538 0.406 77 251.06 0.090 0.0187 0.6545 0.5414 0.0368
TF2 Training 41 0.8739 0.9327 0.8073 0.8587 0.334 0.276 270 251.06 0.048 0.0194 0.8642 pIC50 ¼ 0.4618361 (0.0840195) þ
InvTrain 41 0.8636 0.9201 0.8769 0.8442 0.363 0.263 247 251.06 0.050 0.0292 0.8489 0.0934012 (0.0010513) *
Calib 42 0.7416 0.8584 0.8607 0.7151 0.7487 0.7201 0.8004 0.421 0.323 115 251.14 0.074 0.0086 0.7373 0.6390 0.0339 DCW(4,11)
Valid 41 0.7223 0.8392 0.6313 0.6913 0.4824 0.3706 101 251.06 0.079 0.0189 0.7128 0.6145 0.0059
Comments Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct
8 TF1 Training 41 0.9286 0.9630 0.9177 0.9187 0.251 0.193 507 251.06 0.035 0.0447 0.9059 pIC50 ¼ 0.7891190 (0.0833608) þ
InvTrain 41 0.9285 0.9633 0.9322 0.9206 0.254 0.205 507 251.06 0.035 0.0252 0.9158 0.1757479 (0.0016300) * DCW(7,2)
Calib 41 0.7007 0.8319 0.6708 0.6714 0.6699 0.6416 0.7171 0.501 0.355 91 251.06 0.082 0.0056 0.6979 0.5864 0.0655
Valid 42 0.6556 0.8067 0.4541 0.6096 0.482 0.354 76 251.14 0.091 0.0267 0.6421 0.5302 0.0801
TF2 Training 41 0.8436 0.9151 0.7188 0.8224 0.372 0.289 210 251.06 0.055 0.0119 0.8376 pIC50 ¼ 1.5466503 (0.0956255) þ
InvTrain 41 0.9166 0.9388 0.633 0.9076 0.327 0.277 429 251.06 0.038 0.0379 0.8974 0.1310703 (0.0018611) * DCW(3,5)
6
Calib 41 0.7398 0.8588 0.8564 0.7130 0.7331 0.7102 0.7713 0.450 0.343 111 251.06 0.075 0.0461 0.7164 0.6366 0.0434
Valid 42 0.7437 0.8615 0.6794 0.7136 0.411 0.301 116 251.14 0.073 0.0307 0.7282 0.6415 0.0464
Comments Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Correct Incorrect Correct Correct Correct
Fig. 2. Graphical representation of experimental pIC50 versus calculated pIC50 for eight models developed by TF2.
Table 5
The statistical characteristics of models for CB1 receptor inhibitors for eight random splits.
Split TF1 TF2
7
P. Kumar, A. Kumar Chemometrics and Intelligent Laboratory Systems 200 (2020) 103982
Table 6
Promoters of increase/decrease for endpoint according to selected "depth of interpretation".
S. SAk CWs Probe CWs Probe CWs Probe N1a N2a N3a Defect Comments
No. 1 2 3 [SAk]b
8
P. Kumar, A. Kumar
Table 7
Newly Designed molecules from CB018 and Lipinski’s rule of five.
S. Promoter of endpoint increase ID of designed SMILES Notation Structure calculated Lipinski’s rule of five
No. Compound pIC50
Molecular Number of Number of Virtual Lipinski
Weight Hydrogen Hydrogen logP Rule of
Bond acceptors Bond donors Five
CB018 (Lead Cc1c (nn (c2ccc (Cl)cc2Cl)c1c3ccc 6.9737 464.7721 3 1 6.396 75%
compound) (Cl)cc3)C (¼O)NC(¼O)C(C) (C)C
Combination of sp3 carbon with NCB01 CC(C)c1c (nn (c2ccc (Cl)cc2Cl) 7.1071 492.8253 3 1 7.035 75%
branching c1c3ccc (Cl)cc3)C (¼O)NC(¼O)
C(C) (C)C
Presence of chlorine NCB02 Cc1c (nn (c2ccc (Cl)cc2Cl)c1c3c 7.0819 499.2172 3 1 7.075 75%
(Cl)cc (Cl)cc3)C (¼O)NC(¼O)
C(C) (C)C
9
Combination of chlorine and sp3 NCB03 CC(C)c1c (nn (c2ccc (Cl)cc2Cl) 7.2153 527.2703 3 1 7.698 50%
carbon with branching c1c3c (Cl)cc (Cl)cc3)C (¼O)
NC(¼O)C(C) (C)C
6. Aromatic ring and sp3 carbon with NCB05 CC(C)Cc1c (nn (c2ccc (Cl)cc2Cl) 7.9140 556.9105 3 1 8.318 50%
branching c1c3ccc (Cl)c4ccccc34)C (¼O)
NC(¼O)C(C) (C)C
9 FDA Approved Drug Otenabant [66] CCNC1(CCN(CC1)c1c2nc (n 9.0297 510.4183 2 7 5.3793 50%
(c2ncn1)c1ccc (Cl)cc1) (Expt
c1ccccc1Cl)C (¼O)N 9.1549)
10
Fig. 4. Binding Interactions of newly designed compounds NCB01-NCB07 with lead molecule CB018.
[32]. The IIC was considered as a final parameter to validate the devel- model. The results of CCI are depicted in Table 5 and it shows that
oped QSAR model and all models were correctly predicted by IIC. The smaller the absolute value of CCI, better is the model (more R2valid value).
literature survey also reveals that “the more IIC, the higher statistical The absolute value of CCI for the QSAR model developed with TF2 is less
quality of a model for external validation set” [19,29,50,51]. Hence it can than the QSAR model developed with TF1 (Table 5, Fig. 3). Literature
be generalized that "the IIC is a consistent criterion of predictive potential survey reveals that CCI is more correlated with R2 of the validation set in
of a model". comparison to IIC and in the present manuscript, this outcome is also
The mathematical equation (9) is applied to calculate CCI and it can found true [52]. The absolute value of the correlation of R2valid with
be employed as a criterion of predictive potential of a developed QSAR CCIcalib (R ¼ 0.2013) is more than the correlation with IICcalib (R ¼
11
P. Kumar, A. Kumar Chemometrics and Intelligent Laboratory Systems 200 (2020) 103982
0.0809). Hence we may conclude that CCI has the predictive potential for interactions with active site residues as compared to the methyl group in
the models developed with IIC and it can be used as a statistical criterion CB0018. Additional chloro group in compound NCB02 exhibited hy-
to predict a better model. However, this statement has been tested only drophobic interactions with Leu359, Phe379 and Ser383. Presence of
for the present eight split and this assumption should be verified in future both of these attributes in compound NCB03 headed towards more
with different data set. number of interactions. Increase in the number of aromatic rings as in
An important criterion of OECD principles is the applicability domain compound NCB04 resulted in an increase in the number of pi-pi in-
(AD). Compounds falling outside the applicability domain are defined as teractions (pi-pi stacked and T-shaped) which caused the higher activity
outliers. In case of TF1, without taking into account the influence of IIC of this molecule. However, changing the position of the aromatic ring
on activity (pIC50), the number of outliers for splits 1, 2, 3, 4, 5, 6, 7 and 8 (NBC05) led to a decrease in the activity but it was still higher than the
are 0, 10, 0, 0, 0, 21, 0 and 14 respectively. On the other hand, In case of base compound due to more number of interactions. Addition of nitrogen
TF2, the number of outlier are 0, 8, 17, 11, 22, 26, 16 and 19 for splits 1, with a double bond (molecule NCB06) also caused a better fit of the
2, 3, 4, 5, 6, 7 and 8 respectively by considering the influence of IIC on the molecule as compared to CB018 but it was less active than compound
pIC50 (see supplementary information). NCB04. Chloro group of this compound also exhibited one halogen bond
with Gly166. Combination of oxygen with a double bond, aromatic
3.1. Mechanistic interpretation, designing of the new molecules and system, nitrogen with double bond and presence of at least two aromatic
prediction of market drugs ring formed compound NBC07 which came out to be most active due to
these attributes. Pyrazole ring of this compound displayed pi-sulfur
A mechanistic interpretation is another significant criterion given by interaction with Met384 while iminophenyl ring made pi-donor
OECD, which means that molecular features are accountable for the in- hydrogen bond with NH of Met103. These additional engagements
crease and decrease of an endpoint can be extracted from such models. In might be the cause of the higher activity of this molecule. Except for
Monte Carlo optimization method, the optimal descriptor, correlation compound NCB02 and NCB05, all other molecules also exhibited
weight values (CW), can be correlated with appropriate structural attri- T-shaped pi-pi interaction with Phe268 (Fig. 3). The residues involved in
butes (SAk) which aid in the detection of the promoter of increase and binding interactions with these designed molecules are also involved in
decrease of the endpoint. binding with known antagonists namely rimonabant, otenabant, and
The following three rules are generally applied to categories the taranabant [70].
structural attributes achieved from three independent Monte Carlo
optimization runs as promoter of increase or decrease for endpoint: (i) 4. Conclusion
the structure attribute is promotor of increase if CW(SAk) is positive in all
runs (ii) the structure attribute is promotor of decrease if CW(SAk) is Here we report a detailed QSAR analysis for 165 CB1 receptor in-
negative in all runs (iii) if CW(SAk) is both positive and negative in all hibitors employing the Monte Carlo optimization process incorporated
runs, then that structure attribute is undefined. The list of all structural within the CORAL software. This method is applied to develop 16 models
attributes with their correlation weights for three runs of the developed for eight different random splits using two target function TF1 (without
QSAR model for split 5 by using TF2 is depicted in Table 6. Seven new IIC) and TF2 (with IIC). All the models developed with TF2 have better
cannabinoid CB1 receptor inhibitors (NCB01-NCB07) were designed predictive potential than the models developed with TF1. The model built
from the lead compound CB018. All the newly designed molecules with for split 5 using TF2 is the leading model due to the higher value of the
their chemical structure, SMILES notation, predicted pIC50 and promotor determination coefficient of the validation set (R2Valid ¼ 0.8518). The IIC
of endpoint increase is given in Table 7. The range of predicted pIC50 for improves the statistical performance of CORAL-based QSAR-models and
all the newly designed compounds is 7.0819–9.5048. The predicted gives statistically robust predictive models of the investigated endpoint
pIC50 of the most active designed compound is 9.5048. The Lipinski,s pIC50. In the present manuscript, a novel criterion Correlation Contra-
rule of 5 was also applied to the designed molecules. The compound diction Index is also applied to know its predictive potential. The abso-
CB01 and CB02 follows 75% Lipinski,s rule of 5, whereas for other lute value of CCI for calibration set is less when QSAR models are
compounds follows 50% Lipinski,s rule [65] of 5. In order to check the developed employing IIC. The promoters of increase and decrease
predictive potential of the model, we determined the endpoints of a three endpoint pIC50 are identified and these are applied to design seven new
market drugs such as Otenabant [66], Surinabant [67] and Ibipinament compounds (NCB01-NCB07) and these are docked in the active site of
(SLV319) [68]. The endpoint calculated for these drugs are very close to human cannabinoid receptor CB1 (PDB ID: 5tgz). The binding confor-
the experimental endpoint (Table 7; entry 9–11) mations of compounds (NCB01-NCB07) were stabilized by hydrophobic
interactions mainly. Dichlorophenyl ring was detected in the slender side
3.2. Molecular docking studies pocket and made T shaped pi-pi interactions with Phe170 which is a very
important interaction for the activity of compounds. The residues
To further validate the design strategy, docking studies of base involved in binding interactions with these designed molecules are also
compound CB018 and designed compounds was carried out with the help involved in binding with known antagonists namely rimonabant, ote-
of Autodock Vina program [69] and results were visualized with Dis- nabant, and taranabant [70].
covery Studio Visualizer. The 3D structure of molecules was prepared in
MarvinSketch and CB1 PDB (pdb id: 5tgz) was retrieved from Protein Declaration of competing interest
Database (https://www.rcsb.org/structure/5TGZ) [70]. The compounds
were docked in the active site of human cannabinoid receptor CB1 (PDB The authors declare that they have no known competing financial
ID: 5tgz) and all the compounds emulated the extended anchoring interests or personal relationships that could have appeared to influence
conformation of co-crystalized ligand AM6538. These compounds were the work reported in this paper.
affixed pretty low in the active site of receptor nearly above Trp356 and
capped with Met103 (Fig. 4). The binding conformations of compounds CRediT authorship contribution statement
were stabilized by hydrophobic interactions mainly. Dichlorophenyl ring
was detected in the slender side pocket and made T shaped pi-pi in- Parvin Kumar: Conceptualization, Data curation, Methodology,
teractions with Phe170 which is a very important interaction for the Software, Writing - original draft, Writing - review & editing. Ashwani
activity of compounds [70]. Addition of positive structural attributes Kumar: Conceptualization, Data curation, Formal analysis, Investiga-
resulted in the better fitting of compounds by making suitable in- tion, Methodology, Software, Validation, Visualization, Writing - review
teractions. In compound NCB01, isopropyl group led to more pi alkyl
12
P. Kumar, A. Kumar Chemometrics and Intelligent Laboratory Systems 200 (2020) 103982
& editing. pyrophosphate synthase inhibitors based on Monte Carlo method, Drug Res. 69
(2019) 159–167.
[21] A. Kumar, E. Rathi, S.G. Kini, Identification of potential tumour-associated carbonic
Acknowledgements anhydrase isozyme IX inhibitors: atom-based 3D-QSAR modelling, pharmacophore-
based virtual screening and molecular docking studies, J. Biomol. Struct. Dyn.
The authors are thankful to Dr Andrey A. Toropov and Dr Alla P. (2019), https://doi.org/10.1080/07391102.2019.1626285.
[22] M. Marzo, G.J. Lavado, F. Como, A.P. Toropova, A.A. Toropov, D. Baderna,
Toropova for providing CORAL software. The authors are also thankful to C. Cappelli, A. Lombardo, C. Toma, M. Blazquez, E. Benfenati, QSAR models for
their respective universities for providing the infrastructure. biocides: the example of the prediction of Daphnia magna acute toxicity, SAR QSAR
Environ. Res. 31 (2020) 227–243.
[23] A.A. Toropov, A.P. Toropova, The Monte Carlo Method as a tool to build up
Appendix A. Supplementary data predictive QSPR/QSAR, Curr. Comput. Aided Drug Des. (2019), https://doi.org/
10.2174/1573409915666190328123112.
Supplementary data to this article can be found online at https:// [24] A.P. Toropova, A.A. Toropov, CORAL: QSAR models for carcinogenicity of organic
compounds for male and female rats, Comput. Biol. Chem. 72 (2018) 26–32.
doi.org/10.1016/j.chemolab.2020.103982. [25] P. Kumar, A. Kumar, Monte Carlo method based QSAR studies of mer kinase
inhibitors in compliance with OECD principles, Drug Res. 68 (2018) 189–195.
References [26] A.P. Toropova, A.A. Toropov, E. Carnesecchi, E. Benfenati, J.L. Dorne, The index of
ideality of correlation: models for flammability of binary liquid mixtures, Chem.
Pap. 74 (2020) 601–609.
[1] F. Rossi, F. Punzo, G. Umano, M. Argenziano, E. Miraglia Del Giudice, Role of
[27] S. Jain, S.A. Amin, N. Adhikari, T. Jha, S. Gayen, Good and bad molecular
cannabinoids in obesity, Int. J. Mol. Sci. 19 (2018) 2690.
fingerprints for human rhinovirus 3C protease inhibition: identification, validation,
[2] A. Tremblay, E. Doucet, Obesity: a disease or a biological adaptation? Obes. Rev. 1
and application in designing of new inhibitors through Monte Carlo-based QSAR
(2000) 27–35.
study, J. Biomol. Struct. Dyn. 38 (2020) 66–77.
[3] G.A. Bray, Obesity: the disease, J. Med. Chem. 49 (2006) 4001–4007.
[28] A.P. Toropova, A.A. Toropov, A.M. Veselinovic, J.B. Veselinovic, D. Leszczynska,
[4] P.N. Patel, R. Pathak, Rimonabant: a novel selective cannabinoid-1 receptor
J. Leszczynski, Semi-correlations combined with the index of ideality of correlation:
antagonist for treatment of obesity, Am. J. Health Syst. Pharm. 64 (2007) 481–489.
a tool to build up model of mutagenic potential, Mol. Cell. Biochem. 452 (2019)
[5] D.R. Hou, S. Alam, T.C. Kuan, M. Ramanathan, T.P. Lin, M.S. Hung, 1,2,3-Triazole
133–140.
derivatives as new cannabinoid CB1 receptor antagonists, Bioorg. Med. Chem. Lett
[29] A.P. Toropova, A.A. Toropov, The index of ideality of correlation: improvement of
19 (2009) 1022–1025.
models for toxicity to algae, Nat. Prod. Res. 33 (15) (2019) 2200–2207, https://
[6] H.J. Seo, M.J. Kim, S.H. Lee, S.-H. Lee, M.E. Jung, M.-S. Kim, K. Ahn, J. Kim, J. Lee,
doi.org/10.1080/14786419.2018.1493591.
Synthesis and structure–activity relationship of 1,2,4-triazole-containing
[30] A.A. Toropov, A.P. Toropova, Predicting cytotoxicity of 2-phenylindole derivatives
diarylpyrazolyl carboxamide as CB1 cannabinoid receptor–ligand, Bioorg. Med.
against breast cancer cells using index of ideality of correlation, Anticancer Res. 38
Chem. 18 (2010) 1149–1162.
(2018) 6189–6194.
[7] K.A. Horton, A.V. Goonawardena, J. Sesay, A.C. Howlett, R.E. Hampson, Systemic
[31] S. Ahmadi, Mathematical modeling of cytotoxicity of metal oxide nanoparticles
blockade of the CB1 receptor augments hippocampal gene expression involved in
using the index of ideality correlation criteria, Chemosphere 242 (2020) 125192.
synaptic plasticity but perturbs hippocampus-dependent learning task, Cannabis
[32] P. Kumar, A. Kumar, J. Sindhu, In silico design of diacylglycerol acyltransferase-1
Cannabinoid Res. 4 (2019) 33–41.
(DGAT1) inhibitors based on SMILES descriptors using Monte-Carlo method,, SAR
[8] S.D. Banister, K. Krishna Kumar, V. Kumar, B.K. Kobilka, S.V. Malhotra, Selective
QSAR Environ. Res. 30 (2019) 525–541, https://doi.org/10.1080/
modulation of the cannabinoid type 1 (CB1) receptor as an emerging platform for
1062936X.2019.1629998.
the treatment of neuropathic pain, Medchemcomm 10 (2019) 647–659.
[33] Manisha, S. Chauhan, P. Kumar, A. Kumar, Development of prediction model for
[9] K. Yoshida, Y. Kita, S.M. Tokuoka, F. Hamano, M. Yamazaki, K. Sakimura, M. Kano,
fructose- 1,6- bisphosphatase inhibitors using the Monte Carlo method, SAR QSAR
T. Shimizu, Monoacylglycerol lipase deficiency affects diet-induced obesity, fat
Environ. Res. 30 (2019) 145–159.
absorption, and feeding behavior in CB1 cannabinoid receptor-deficient mice,
[34] A. Kumar, S. Singh, S. Jain, P. Kumar, Synthesis, antimicrobial evaluation, QSAR
Faseb. J. 33 (2019) 2484–2497.
and in Silico ADMET studies of decanoic acid derivatives, Acta Pol. Pharm. 68
[10] X. Xu, S. Jiang, E. Xu, X. Wu, R. Zhao, Inhibition of CB1 receptor ameliorates spatial
(2011) 191–204.
learning and memory impairment in mice with traumatic brain injury, Neurosci.
[35] P. Kumar, M. Duhan, K. Kadyan, J.K. Bhardwaj, P. Saraf, M. Mittal, Multicomponent
Lett. 696 (2019) 127–131.
synthesis of some molecular hybrid containing thiazole pyrazole as apoptosis
[11] M.C.G. Leite-Avalca, F.T. Staats, D. Verona, P. de Souza, M.C. Almeida, J.E. Silva-
inducer, Drug Res. 68 (2018) 72–79.
Santos, A.R. Zampronio, Cannabinoid CB1 receptor antagonist rimonabant
[36] P. Kumar, K. Kadyan, M. Duhan, J. Sindhu, V. Singh, B.S. Saharan, Design,
decreases levels of markers of organ dysfunction and alters vascular reactivity in
synthesis, conformational and molecular docking study of some novel acyl
aortic vessels in late sepsis in rats, Inflammation 42 (2019) 618–627.
hydrazone based molecular hybrids as antimalarial and antimicrobial agents, Chem.
[12] J.L.C. Lee, F.E. Amorim, L.F. Cassini, O.B. Amaral, Different temporal windows for
Cent. J. 11 (2017) 115.
CB1 receptor involvement in contextual fear memory destabilisation in the
[37] P. Kumar, M. Duhan, K. Kadyan, J. Sindhu, S. Kumar, H. Sharma, Synthesis of novel
amygdala and hippocampus, PloS One 14 (2019), e0205781.
inhibitors of α-amylase based on thiazolidine-4-one skeleton containing pyrazole
[13] H. Laurikainen, L. Tuominen, M. Tikka, H. Merisaari, R.L. Armio, E. Sormunen,
moiety and their configurational studies, MedChemComm 8 (2017) 1468–1476,
F. Borgan, M. Veronese, O. Howes, M. Haaparanta-Solin, O. Solin, J. Hietala,
https://doi.org/10.1039/C7MD00080D.
M. group, Sex difference in brain CB1 receptor availability in man, Neuroimage 184
[38] P. Kumar, R. Bhatia, R. Khanna, A. Dalal, D. Kumar, P. Surain, R.C. Kamboj,
(2019) 834–842.
Synthesis of some benzothiazoles by developing a new protocol using urea nitrate
[14] B.K. Srivastava, R. Soni, A. Joharapurkar, K.V. Sairam, J.Z. Patel, A. Goswami,
as a catalyst and their antimicrobial activities, J. Sulfur Chem. 38 (2017) 585–596.
S.A. Shedage, S.S. Kar, R.P. Salunke, S.B. Gugale, A. Dhawas, P. Kadam, B. Mishra,
[39] P. Kumar, R. Bhatia, D. Kumar, R.C. Kamboj, S. Kumar, R. Kamal, R. Kumar, An
N. Sadhwani, V.B. Unadkat, P. Mitra, M.R. Jain, P.R. Patel, Bioisosteric replacement
economic, simple and convenient synthesis of 2-aryl/heteroaryl/styryl/alkylben-
of dihydropyrazole of 4S-(-)-3-(4-chlorophenyl)-N-methyl-N’-[(4-chlorophenyl)-
zothiazoles using SiO 2–HNO 3, Res. Chem. Intermed. 41 (2015) 4283–4292.
sulfonyl]-4-phenyl-4,5-di hydro-1H-pyrazole-1-caboxamidine (SLV-319) a potent
[40] S. Kumar, S. Kumar, P. Kumar, Synthesis and antimicrobial activity of some (3-
CB1 receptor antagonist by imidazole and oxazole, Bioorg. Med. Chem. Lett 18
phenyl-5-(1-phenyl-3-aryl-1H-pyrazol-4-yl)-4, 5-dihydro-1H-pyrazol-1-yl)(pyridin-
(2008) 963–968.
4-yl) methanones: new derivatives of 1, 3, 5-trisubstituted pyrazolines, Med. Chem.
[15] C. Montero, N.E. Campillo, P. Goya, J.A. Paez, Homology models of the cannabinoid
Res. 22 (2013) 433–439.
CB1 and CB2 receptors. A docking analysis study, Eur. J. Med. Chem. 40 (2005)
[41] P. Kumar, A. Kumar, J. Makrandi, Synthesis and evaluation of bioactivity of
75–83.
thiazolo [3, 2-b]-[1, 2, 4]-triazoles and isomeric thiazolo [2, 3-c]-[1, 2, 4]-triazoles,
[16] C.-J. Qiao, H.I. Ali, K.H. Ahn, S. Kolluru, D.A. Kendall, D. Lu, Synthesis and
J. Heterocycl. Chem. 50 (2013) 1223–1229.
biological evaluation of indole-2-carboxamides bearing photoactivatable
[42] P. Kumar, A. Kuamr, L.J. Mohan, J. Makrandi, Heterocyclic systems containing
functionalities as novel allosteric modulators for the cannabinoid CB1 receptor, Eur.
bridgehead nitrogen atom: synthesis and evaluation of biological activity of
J. Med. Chem. 121 (2016) 517–529.
imidazo [2, 1-b]-1, 3, 4-thiadiazolo [2, 3-c]-s-triazoles, s-triazolo [3, 4-b]-1, 3, 4-
[17] J. Lee, H.J. Seo, S.H. Lee, J. Kim, M.E. Jung, S.-H. Lee, K.-S. Song, J. Lee, S.Y. Kang,
thiadiazolo [3, 2-b] imidazo [4, 5-b] quinoxaline and bis-(s-Triazolo [3, 4-b]-1, 3, 4-
M.J. Kim, M.-S. Kim, E.-J. Son, M. Lee, H.-K. Han, Discovery of 2-(4-((1H-1,2,4-
thiadiazolo [3, 2-b][imidazo [4, 5-b]-cyclohexane]-5a, 6a-diene), Bull. Kor. Chem.
triazol-1-yl)methyl)-5-(4-bromophenyl)-1-(2-chlorophenyl)-1H-pyrazol-3-yl)-5-
Soc. 31 (2010) 3304–3308.
tert-butyl-1,3,4-thiadiazole (GCC2680) as a potent, selective and orally efficacious
[43] K.-S. Song, M.J. Kim, H.J. Seo, S.-H. Lee, M.E. Jung, S.-U. Kim, J. Kim, J. Lee,
cannabinoid-1 receptor antagonist, Bioorg. Med. Chem. 18 (2010) 6377–6388.
Synthesis and structure–activity relationship of novel diarylpyrazole imide
[18] M.K. Sharma, P.R. Murumkar, A.M. Kanhed, R. Giridhar, M.R. Yadav, Prospective
analogues as CB1 cannabinoid receptor ligands, Bioorg. Med. Chem. 17 (2009)
therapeutic agents for obesity: molecular modification approaches of centrally and
3080–3092.
peripherally acting selective cannabinoid 1 receptor antagonists, Eur. J. Med.
[44] N. Khan, S.A. Halim, W. Khan, S.K. Zafar, Z. Ul-Haq, In-silico designing and
Chem. 79 (2014) 298–339.
characterization of binding modes of two novel inhibitors for CB1 receptor against
[19] P. Kumar, A. Kumar, J. Sindhu, Design and development of novel focal adhesion
obesity by classical 3D-QSAR approach, J. Mol. Graph. Model. 89 (2019) 199–214.
kinase (FAK) inhibitors using Monte Carlo method with index of ideality of
[45] A.A. Toropov, A.P. Toropova, E. Benfenati, Additive SMILES-based carcinogenicity
correlation to validate QSAR, SAR QSAR Environ. Res. 30 (2019) 63–80.
models: probabilistic principles in the search for robust predictions, Int. J. Mol. Sci.
[20] P. Kumar, A. Kumar, J. Sindhu, S. Lal, QSAR models for nitrogen containing
10 (2009) 3106–3127.
monophosphonate and bisphosphonate derivatives as human farnesyl
13
P. Kumar, A. Kumar Chemometrics and Intelligent Laboratory Systems 200 (2020) 103982
[46] A.P. Toropova, A.A. Toropov, D. Leszczynska, J. Leszczynski, CORAL and Nano- [60] P. Gramatica, A. Sangion, A historical excursus on the statistical validation
QFAR: quantitative feature - activity relationships (QFAR) for bioavailability of parameters for QSAR models: a clarification concerning metrics and terminology,
nanoparticles (ZnO, CuO, Co3O4, and TiO2), Ecotoxicol. Environ. Saf. 139 (2017) J. Chem. Inf. Model. 56 (2016) 1127–1131.
404–407. [61] P. Kumar, A. Kumar, Nucleobase sequence based building up of reliable QSAR
[47] P.G.R. Achary, A.P. Toropova, A.A. Toropov, Combinations of graph invariants and models with the index of ideality correlation using Monte Carlo method, J. Biomol.
attributes of simplified molecular input-line entry system (SMILES) to build up Struct. Dyn. (2019), https://doi.org/10.1080/07391102.2019.1656109.
models for sweetness, Food Res. Int. 122 (2019) 40–46. [62] P. Kumar, M. Nimbhal, K. Bagri, P. Kumar, The index of ideality of correlation: A
[48] M. Duhan, R. Singh, M. Devi, J. Sindhu, R. Bhatia, A. Kumar, P. Kumar, Synthesis, statistical yardstick for better QSAR modeling of glucokinase activators, Struct.
molecular docking and QSAR study of thiazole clubbed pyrazole hybrid as Chem. (2019), https://doi.org/10.1007/s11224-019-01468-w.
α-amylase inhibitor, J. Biomol. Struct. Dyn. (2019), https://doi.org/10.1080/ [63] P.K. Ojha, I. Mitra, R.N. Das, K. Roy, Further exploring rm2 metrics for validation of
07391102.2019.17. QSPR models, Chemometr. Intell. Lab 107 (2011) 194–205.
[49] A.P. Toropova, A.A. Toropov, QSPR and nano-QSPR: what is the difference? J. Mol. [64] A. Rescifina, G. Floresta, A. Marrazzo, C. Parenti, O. Prezzavento, G. Nastasi,
Struct. 1182 (2019) 141–149. M. Dichiara, E. Amata, Development of a Sigma-2 Receptor affinity filter through a
[50] A.P. Toropova, A.A. Toropov, The index of ideality of correlation: a criterion of Monte Carlo based QSAR analysis, Eur. J. Pharmaceut. Sci. 106 (2017) 94–101.
predictability of QSAR models for skin permeability? Sci. Total Environ. 586 (2017) [65] C.A. Lipinski, F. Lombardo, B.W. Dominy, P.J. Feeney, Experimental and
466–472. computational approaches to estimate solubility and permeability in drug discovery
[51] A.A. Toropov, A.P. Toropova, Use of the index of ideality of correlation to improve and development settings1PII of original article: S0169-409X(96)00423-1. The
predictive potential for biochemical endpoints, Toxicol. Mech. Methods 29 (2019) article was originally published in Advanced Drug Delivery Reviews 23 (1997)
43–52. 3–25.1, Adv. Drug Deliv. Rev. 46 (2001) 3–26.
[52] A.A. Toropov, A.P. Toropova, The Correlation Contradictions Index (CCI): building [66] D.A. Griffith, J.R. Hadcock, S.C. Black, P.A. Iredale, P.A. Carpino, P. DaSilva-
up reliable models of mutagenic potential of silver nanoparticles under different Jardine, R. Day, J. DiBrino, R.L. Dow, M.S. Landis, R.E. O’Connor, D.O. Scott,
conditions using quasi-SMILES, Sci. Total Environ. 681 (2019) 102–109. Discovery of 1-[9-(4-Chlorophenyl)-8-(2-chlorophenyl)-9H-purin-6-yl]-4-ethylami-
[53] R.B. Aher, K. Roy, Exploring the structural requirements in multiple chemical nopiperidine-4-carboxylic acid amide hydrochloride (CP-945,598), a novel, potent,
scaffolds for the selective inhibition of Plasmodium falciparum calcium-dependent and selective cannabinoid type 1 receptor antagonist, J. Med. Chem. 52 (2009)
protein kinase-1 (PfCDPK-1) by 3D-pharmacophore modelling, and docking studies, 234–237.
SAR QSAR Environ. Res. 28 (2017) 390–414. [67] L.E. Klumpers, C. Roy, G. Ferron, S. Turpault, F. Poitiers, J.-L. Pinquier, J.G.C. van
[54] S.S. Bhayye, K. Roy, A. Saha, Pharmacophore generation, atom-based 3D-QSAR, Hasselt, L. Zuurman, F.A.S. Erwich, J.M.A. van Gerven, Surinabant, a selective
HQSAR and activity cliff analyses of benzothiazine and deazaxanthine derivatives cannabinoid receptor type 1 antagonist, inhibits Δ9-tetrahydrocannabinol-induced
as dual A2A antagonists/MAOB inhibitors,, SAR QSAR Environ. Res. 27 (3) (2016) central nervous system and heart rate effects in humans, Br. J. Clin. Pharmacol. 76
183–202, https://doi.org/10.1080/1062936X.2015.1136840. (2013) 65–77.
[55] A. Golbraikh, A. Tropsha, Beware of q2!, J. Mol. Graph. Model. 20 (2002) 269–276. [68] R.J. Chorvat, J. Berbaum, K. Seriacki, J.F. McElroy, JD-5006 and JD-5037:
[56] L.M. Shi, H. Fang, W. Tong, J. Wu, R. Perkins, R.M. Blair, W.S. Branham, S.L. Dial, peripherally restricted (PR) cannabinoid-1 receptor blockers related to SLV-319
C.L. Moland, D.M. Sheehan, QSAR models using a large diverse set of estrogens, (Ibipinabant) as metabolic disorder therapeutics devoid of CNS liabilities, Bioorg.
J. Chem. Inf. Comput. Sci. 41 (2001) 186–195. Med. Chem. Lett 22 (2012) 6173–6180.
[57] G. Schüürmann, R.-U. Ebert, J. Chen, B. Wang, R. Kühne, External validation and [69] O. Trott, A.J. Olson, AutoDock Vina, Improving the speed and accuracy of docking
prediction employing the predictive squared correlation coefficient — test set with a new scoring function, efficient optimization, and multithreading, J. Comput.
activity mean vs training set activity mean, J. Chem. Inf. Model. 48 (2008) Chem. 31 (2010) 455–461.
2140–2145. [70] T. Hua, K. Vemuri, M. Pu, L. Qu, G.W. Han, Y. Wu, S. Zhao, W. Shui, S. Li, A. Korde,
[58] P. Pratim Roy, S. Paul, I. Mitra, K. Roy, On two novel parameters for validation of R.B. Laprairie, E.L. Stahl, J.H. Ho, N. Zvonok, H. Zhou, I. Kufareva, B. Wu, Q. Zhao,
predictive QSAR models, Molecules 14 (2009) 1660–1701. M.A. Hanson, L.M. Bohn, A. Makriyannis, R.C. Stevens, Z.J. Liu, Crystal structure of
[59] K. Roy, P. Chakraborty, I. Mitra, P.K. Ojha, S. Kar, R.N. Das, Some case studies on the human cannabinoid receptor CB1, Cell 167 (2016) 750–762, https://doi.org/
application of “rm2” metrics for judging quality of quantitative structure–activity 10.1016/j.cell.2016.10.004, e14.
relationship predictions: emphasis on scaling of response data, J. Comput. Chem. 34
(2013) 1071–1082.
14