Heliyon 8 (2022) e11537

Contents lists available at ScienceDirect

Heliyon
journal homepage: www.cell.com/heliyon

Research article

In silico design of EGFRL858R/T790M/C797S inhibitors via 3D-QSAR, molecular

docking, ADMET properties and molecular dynamics simulations
Hanine Hadni ∗ , Menana Elhallaouia
LIMAS, Faculty of Sciences Dhar El Mahraz, Sidi Mohamed Ben Abdellah University, Fez, Morocco

A R T I C L E I N F O A B S T R A C T

Keywords: The development of L858R/T790M/C797S mutations in EGFR is one of the main reasons for the emergence of
Non-small cell lung cancer resistance after third-generation treatment of non-small cell lung cancer (NSCLC). Therefore, the development
3D-QSAR of 4th generation drugs needs urgent attention. To overcome resistance, in silico drug discovery and Design
EGFRL858R/T790M/C797S inhibition
approaches were employed on a library of 29 novel 9-heterocyclyl substituted 9H-purines derivatives with
ADMET
Molecular docking
EGFRL858R/T790M/C797S inhibition for anticancer activity against NSCLC. The COMSIA/EHA model (Q2 = 0.584, R2
Molecular dynamics simulation = 0.816, and 𝑅2𝑝𝑟𝑒𝑑 = 0.73) showed a stable and reliable predictive ability of NSCLC activity, which was tested
by several validation methods. Molecular docking studies reveal crucial interactions with EGFRL858R/T790M/C797S
inhibition for NSCLC activity. Based on theoretical methods, we designed 10 new compounds with good activity
potential, which were tested using ADMET properties. Next, the molecular docking results were examined by
molecular dynamics simulations to verify the stability of hydrogen bonding interactions with important residues
such as MET790, MET793 and SER797, which are essential for the design of 4th generation EGFR Inhibitors to
combat drug-resistant NSCLC.

1. Introduction mutation, second-generation EGFR-TKIs, namely Afatinib and Dacomi-

Lung cancer is responsible for the highest cancer mortality in men
and women worldwide [1]. According to World Health Organization
(WHO) estimates in 2020 [2], the number of lung cancer cases is ex-
pected to exceed 2.21 million and cause 1.80 million deaths worldwide.
Thus, the cure rate for lung cancer is less than 20%, indicating that
we need to strengthen our efforts to fight lung cancer. Non-small cell
lung cancer (NSCLC) accounts for 85% of lung cancers [3]. The Epider-
mal Growth Factor Receptor (EGFR) is a receptor tyrosine kinase of the
ErbB family, which plays a key role in the transmission of cell signalling
that leads to proliferation, differentiation, apoptosis and migration [4,
5, 6]. The discovery of EGFR tyrosine kinase inhibitors (EGFR-TKIs)
represents an important milestone in NSCLC treatment procedures [7].
However, mutations in the EGFR gene have had a considerable impact
on these procedures, resulting in three generations of EGFR-TKIs.
First-generation EGFR-TKIs (Gefitinib and Erlotinib) have been
shown to be effective in treating NSCLC in patients with the L858R
mutation in EGFR [8, 9, 10]. However, the appearance of the T790M
mutation in the EGFRT gene has led to the therapeutic failure of first-
generation inhibitors [11]. To overcome the resistance to the T790M
tinib, have been developed [12]. However, these inhibitors cause seri-
ous side effects [13]. To overcome the mutations and observed toxicity,
the development of third-generation EGFR-TKIs (rociletinib, osimer-
tinib) avoided these problems with acceptable side effects and improved
action against the T790M mutation [14]. Despite the significant thera-
peutic success of third-generation EGFR-TKIs, the development of new
resistance occurring on target is the C797S mutation, rendering osimer-
tinib clinically ineffective [15]. Therefore, the discovery of new drugs
to overcome the L858R/T790M/C797S mutants of the EGFR represents
an urgent therapeutic need for the treatment of NSCLC.
In recent years, the use of molecular modeling techniques has
yielded very impressive results in the new drug discovery process [16,
17, 18]. For this purpose, three-dimensional quantitative structure–
activity relationship (3D-QSAR), molecular docking, pharmacokinetic
parameters (ADMET) and Molecular dynamic (MD) simulation have
been performed for the design of new molecule able to overcome
the resistance of EGFRL858R/T790M/C797S mutation. In this study, 29
EGFRL858R/T790M/C797S inhibitors were studied using two methods, 3D-
QSAR and molecular docking to identify key structural factors affecting
inhibitory activity. In addition, it helps us to obtain recommendations

* Corresponding author.
E-mail address: hadni.hanine@yahoo.fr (H. Hadni).

https://doi.org/10.1016/j.heliyon.2022.e11537
Received 17 February 2022; Received in revised form 24 May 2022; Accepted 4 November 2022

2405-8440/© 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/li-
censes/by-nc-nd/4.0/).
H. Hadni and M. Elhallaouia Heliyon 8 (2022) e11537

the contributions of each field. In order to test the predictive ability and
stability of the models, several validation strategies were also applied.

2.4. 3D-QSAR models validation

Fig. 1. Structures of 9-heterocyclyl substituted 9H-purines derivatives.
for the design of new drug candidates from the ZINC database. To
test their drug-like potential, all designed compound was evaluated by
calculating the ADMET properties. Furthermore, Molecular dynamics
simulations were performed for 100 ns to estimate the ligand stability
within the protein under normal physiological conditions.
2. Material and methods
2.1. Data set collection and biological activities
2.4.1. Y-randomization
The Y randomization technique is a widely used approach to ensure
the robustness of a 3D-QSAR model [24]. In this test, only the values
of the dependent variable (biological activity) are randomly shuffled
several times, while the values of the independent variable (descriptors)
remain unchanged. These random data are then used to generate new
3D-QSAR models, the 𝑄2𝑦𝑟𝑎𝑛𝑑 and 𝑅2𝑦𝑟𝑎𝑛𝑑 values of the random model are
then tested against the Eriksson and Wold criteria [25] represented by
the following rule:
∙ 𝑄2𝑦𝑟𝑎𝑛𝑑 < 0.2 and 𝑅2𝑦𝑟𝑎𝑛𝑑 < 0.2, no chance correlation
∙ any 𝑄2𝑦𝑟𝑎𝑛𝑑 and 0.2 < 𝑅2𝑦𝑟𝑎𝑛𝑑 < 0.3, negligible chance correlation
∙ any 𝑄2𝑦𝑟𝑎𝑛𝑑 and 0.3 < 𝑅2𝑦𝑟𝑎𝑛𝑑 < 0.4, tolerable chance correlation
∙ any 𝑄2𝑦𝑟𝑎𝑛𝑑 and 𝑅2𝑦𝑟𝑎𝑛𝑑 > 0.4, recognized chance correlation.

An additional criterion based on the parameter called c 𝑅2𝑟 which must

In this work, a series of 29 novel 9-heterocyclyl substituted 9H-
purines derivatives (Fig. 1) as EGFR-TKIs of the L858R/T790M/C797S be greater than 0.5, was calculated according to the following equation:
mutants was collected from the recent literature [19]. Thus, the ob- √
𝑅𝑟 = 𝑅 × (𝑅2 − 𝑅2𝑟 )
c 2
(1)
served activity values (IC50) were converted to their negative loga-
rithm pIC50 (−logIC50) values to build the 3D-QSAR models. To build
construct the 3D-QSAR models (Table 1), the dataset containing 29 2.4.2. External validation of the CoMSIA model
compounds was randomly divided into a training set containing 21 com- To verify the predictive capabilities of the proposed models, the bi-
pounds to generate the 3D-QSAR models (Table 1), while the remaining ological activities of the test set of 8 compounds were predicted. The
8 compounds 5 represented by the symbol “*” superscript were used as predictive ability of the models is measured by the external validation
a test set to verify the performance and accuracy of the established coefficient of determination (𝑅2𝑝𝑟𝑒𝑑 ) calculated by the following formula:
models.
𝑅2𝑝𝑟𝑒𝑑 = (𝑆𝐷 − 𝑃 𝑅𝐸𝑆𝑆)∕𝑆𝐷 (2)
2.2. Structural minimization and alignment The value of SD is the sum of the squared deviations between the
biological activity of the compounds in the test set and the mean activity
Molecular alignment is one of the most important parameters of 3D- of the training set, and the predictive sum of squares (PRESS) is the sum
QSAR studies, as this step is a determining factor of the successful model of the squared deviations between the predicted and actual activities
[20]. All ligands were constructed with the sketch module and opti- of the test set compounds. To test the reliability of 3D-QSAR models in
mized under the Tripos force field, using Gasteiger-Hückel charges [21]. predicting the activity of new compounds, external validation statistical
The convergence criterion of the Powell gradient algorithm was set to criteria was introduced by Golbraikh and Tropsha [26] described by the
0.005 kcal/(mol Å) and a maximum iteration coefficient of 10,000 to following equations:
obtain a stable conformation [22]. All optimized molecular structures
∑
were aligned to the most active compound 21 of the series, using the (Ytest(pred) − kYtest(pred) )2
ALIGN DATABASE method available in SYBYL-X 2.0 software. r02 = 1 − ∑ (3)
(Ytest(pred) − Ytest(pred) )2
∑
2.3. 3D-QSAR studies (Y − kYtest )2
r0′ 2 = 1 − ∑ test (4)
(Ytest − Ytest )2
Ligand-guided alignment was used to derive the CoMSIA models us- ∑
(Ytest − Ytest(pred) )
ing SYBYL-X 2.1. The aligned inhibitors of the training set were placed K= ∑ (5)
inside a 3D cubic lattice with a grid spacing of 2 Å in all Cartesians (Ytest(pred) )2
directions. The CoMSIA descriptors, which includes five force fields ∑
(Ytest − Ytest(pred) )
(steric, electrostatic, hydrophobic, hydrogen bond donors and accep- K′ = ∑ (6)
(Ytest )2
tors), were calculated using a sp3 hybridized carbon probe atom with
a van der Waals radius of 1.52 Å and a net charge of +1.0. An energy Where r02 and r0′ 2 indicate the squared correlation coefficients
cut-off value was set at 30 kcal/mol. The minimum column filtering and through origin between predicted and observed activity, as well as ob-
attenuation factor were set to 2.0 and 0.3 kcal/mol, respectively. served and predicted activity of the test set, respectively. k and k′ are
the slopes for regression through origin.
2.3.1. Partial least squares analysis Furthermore, according to Roy’s statistical criteria [25] to validate
The partial least squares (PLS) analysis was used to build the CoM- the predictive ability of the model, it is important to calculate the dif-
SIA models [23]. First, the leave-one-out (LOO) cross-validation method ference between the values of r2 , 𝑟20 and 𝑟′02 according the following
was applied to obtain the optimal number of components (ONC) with equation:
the highest cross-validation coefficient of determination (Q2 ) to validate √
this step. Second, after determining the ONC, CoMSIA models were built 𝑟2𝑚 = 𝑟2 (1 − (𝑟2 − 𝑟20 ) (7)
based on statistical parameters such as the coefficient of determination √
(R2 ), the standard error of estimate (SEE), the F-value (Fischer test) and 𝑟′𝑚2 = 𝑟2 (1 − (𝑟2 − 𝑟′02 ) (8)

2
H. Hadni and M. Elhallaouia Heliyon 8 (2022) e11537

Table 1. Observed and predicted anticancer activity against NSCLC of 9-heterocyclyl substituted 9H-purines derivatives.
Compounds R1 R2 IC50 (μM) pIC50 (obs) pIC50 (pred) Residual

1 H 0.11 6.958 6.726 0.231

2 F 0.58 6.236 6.124 0.112

3* H 0.012 7.921 7.607 0.313

4 F 0.027 7.568 7.48 0.087

5* F 0.12 6.921 7.222 −0.301

6 F 0.28 6.553 6.709 −0.156

7 H 0.53 6.276 6.919 −0.643

8 F 1.22 5.914 6.332 −0.418

9* H 0.065 7.187 6.875 0.311

10 F 0.3 6.523 6.319 0.204

11* F 0.038 7.421 6.98 0.441

12 F 0.12 6.921 6.784 0.136

13 H 0.17 6.769 6.221 0.547

14 H 0.26 6.585 7.068 −0.483

15* H 0.017 7.769 7.696 0.072

16 H 0.0016 8.796 8.058 0.738

17 F 0.38 6.421 6.534 −0.113

(continued on next page)

3
H. Hadni and M. Elhallaouia Heliyon 8 (2022) e11537

Table 1 (continued)
Compounds R1 R2 IC50 (μM) pIC50 (obs) pIC50 (pred) Residual

18 F 1.52 5.818 5.638 0.179

19 F 1.81 5.742 5.722 0.019

20* F 0.0075 8.125 7.892 0.232

21 H 0.00088 9.055 8.919 0.135

22 F 0.35 6.456 7.092 −0.635

23 F 0.0025 8.602 8.345 0.256

24 F 0.0018 8.745 8.029 0.715

25 F 0.0086 8.065 8.759 −0.694

26* F 0.066 7.181 6.959 0.221

27 F 0.01 8 8.163 −0.163

28 F 0.0075 8.125 8.179 −0.054

29* F 0.05 7.301 7.023 0.278

2.4.3. Applicability domain
The applicability domain (AD) [27] has been applied to determine
the region of chemical space of a 3D QSAR model, which can be used
to reliably predict new compounds. In this study, one of the simple
approaches to determine the DA is the leverage approach. The leverage
values (h) were calculated for each molecule for a graphical detection
of outliers in the Williams diagram by the following relationship:
ℎ𝑖 = (𝑥𝑡𝑖 (𝑋𝑋 𝑡 )−1 𝑥𝑖 ) with (i = 1, 2, … n) (9)
In this equation: xi: the descriptor vector of the tested compound,
X: the matrix n*(k − 1) where n is the number of compounds and k the
number of descriptors in the training set and the exponent t refers to the
matrix/vector transposition. This diagram validates the reliability of the
3D-QSAR model if the value of h is lower than the critical value of the
leverage (h*) (h* = 2.5 (k + 1)/n) [28, 29], where n is the number of
compounds and k is the number of descriptors in the training set. The
AD of the 3D-QSAR model is located to the left of the vertical line of h*
= 0.47.
2.5. Molecular docking investigation
Molecular docking is a reliable computational method, which can be
performed to analyze the interactions between potential drugs and the
active site of the target protein, in order to better understand the key
structural requirements of a geometric model as a function of binding
energy. In the present work, we performed the molecular docking study
of two compounds 4 and 2 with the same R1 and R2 substitution, but
with different R and S enantiomers, with compound 4 having higher
activity than compound 2, in the active site of EGFRL858R/T790M/C797S
(PDB code 6lud), which is downloaded from the RCSB protein database
[30]. Firstly, Discovery Studio software [31] was used to prepare the
protein by removing all water molecules, ligands and non-protein parts.

4
H. Hadni and M. Elhallaouia Heliyon 8 (2022) e11537

Table 2. The PLS statistical results.

Q2 R2 RMSE F ONC 𝑅2𝑝𝑟𝑒𝑑 Fractions

Ster Elec Hyd Don Acc

CoMSIA/SEA 0.584 0.844 0.445 48.687 2 0.603 0.121 0.216 - 0.663
CoMSIA/SEH 0.386 0.816 0.497 25.198 3 0.408 0.249 0.432 0.319 -
CoMSIA/SHA 0.579 0.849 0.437 50.773 3 0.676 0.170 0.132 - 0.698
CoMSIA/EHA 0.571 0.852 0.434 51.812 2 0.715 0.247 0.117 - 0.636
CoMSIA/SEHA 0.567 0.846 0.442 69.893 2 0.680 0.112 0.198 0.108 - 0.582

AutoDock Tools software version 1.5.6 was used to analyze ligand-

protein interactions [32]. The 3D grid was created by the AUTOGRID
algorithm to measure the energies of ligand-protein interactions [33].
The grid maps were generated using 60 Å in all Cartesian directions,
with grid-spacing 0.375 Å, the grid center coordinates are approxi-
mately (−53.330 Å, −5.149 Å, and −17.579 Å) for the location of the
ligand in the complex. The 2D and 3D visualizations and the analysis
of the docked conformations according to the established interactions
were performed using Discovery Studio [31].

2.5.1. Docking validation protocol

The molecular docking simulation was verified by re-docking the
crystallized ligand (drug Osimertinib) into the protein (6lud.pdb). Fig. 2. Alignment of all compounds.
Therefore, the crystallised ligand was extracted from the protein and
docked to the same protein. The lowest energy pose of the docked lig-
Table 3. Statistical parameters of the COMSIA/EHA
and and the crystallised ligand was superimposed, in order to calculate
model after several Y-Randomization tests.
the root mean square deviation (RMSD). To validate the docking pro-
Iteration COMSIA/EHA
cess, the RMSD must be less than 2 Å [34, 35, 36].
𝑄2𝑦𝑟𝑎𝑛𝑑 𝑅2𝑦𝑟𝑎𝑛𝑑 c
𝑅2𝑟
2.6. ADMET prediction studies 1 −0.22 0.340 0.66
2 −0.239 0.388 0.628
3 −0.176 0.330 0.666
The development of computational technology in the pharmaceuti-
4 0.060 0.360 0.647
cal field has made it possible to identify new drug candidates, reduc-
5 −0.143 0.404 0.617
ing the number of experimental tests and improving the success rate.
With this in mind, the in silico study provides a pathway for ADMET
(Absorption, Distribution, Metabolism, Excretion and Toxicity) pharma- The most active compound 20 of the series was chosen as the model
cokinetic parameters [37, 38], with drug absorption in the human gut, molecule to align the data set and to visualise the CoMSIA model con-
drug penetration into the central nervous system and the blood-brain tour maps. Fig. 2 illustrates the minimisation and alignment of all 3D
barrier, metabolism is the chemical biotransformation of a drug by the structures of the data sets to the common core based on compound 21.
body, excretion is the elimination of a drug by the body and the toxicity
levels of a drug. 3.2. CoMSIA studies

2.7. Molecular dynamic simulation studies
Molecular dynamics (MD) simulations were carried out using the
NAMD (Nanoscale Molecular Dynamics) program [39]. We used the
CHARMM-GUI server to generate the NAMD input files [40]. The
CHARMM36 force field was applied to proteins and small molecules.
Protein-ligand systems were solvated with the TIP3P water model in a
10 Å cubic box around and neutralised using the KCl salt at the 0.15 M
ionic concentration [41]. The energy of all systems was minimised for
10,000 steps using the steepest descent method, and equilibrated under
constant atom number, volume and temperature (NVT) at 310 K for 10
ns in an ensemble. Finally, the system was subjected to 100 ns of un-
constrained MD simulations in a constant atomic number, pressure and
temperature (NPT) ensemble with temperature (310 K) and pressure (1
atm) [42]. The analyses of the MD trajectories were used to generate
the root mean square deviation (RMSD), root mean square fluctuation
(RMSF) using Visual Molecular Dynamics (VMD) software [43] to check
the stability of the systems.
3. Results and discussion
3.1. Molecular minimization and alignment
The compound-based molecular alignment method was performed
using the available alignment rule to build a powerful 3D-QSAR model.
The statistical parameters of the CoMSIA models are presented in
Table 2. For the CoMSIA analysis, the combination of the five molecular
fields was used to develop the different CoMSIA models. However, the
hydrogen bond donor field is zero for the molecules in the series. The
results indicate that the best combination was the electrostatic field, the
hydrophobic field and the hydrogen bond acceptor field (EHA), with
The contribution rates were 24.7%, 11.7% and 63.6%, respectively. In
the CoMSIA/EHA model, the Q2 was 0.571 with two as the optimal
number of components, The R2 was 0.852 with a reliable SEE of 0.434
and the F-test value is 51.812, highest predictive value was obtained
𝑅2𝑝𝑟𝑒𝑑 = 0.715 for the external validation of the test set.
Overall, the proposed model is considered a reliable predictive
model, if the Q2 and R2 values are greater than 0.5 and 0.6, respec-
tively, with a prediction value 𝑅2𝑝𝑟𝑒𝑑 of the new compound activity
greater than 0.6. Thus, the CoMSIA/EHA model indicates statistical
significance and good predictive quality, which was confirmed by the
predictive ability of the external validation. For more precision con-
cerning the stability and prediction of the CoMSIA/EHA model, several
validation methods such as the Y-randomization test, the Tropsha and
Golbraikh criteria and Roy criteria were used. Table 3 presents the re-
sults of the Y-randomization test of the CoMSIA/EHA model.
In the results in Table 3, five random mixtures of the Y vector
were performed, according to the criteria of the Y-randomization test
by Eq. (1) described above. The values of 𝑄2𝑦𝑟𝑎𝑛𝑑 , 𝑅2𝑦𝑟𝑎𝑛𝑑 and c 𝑅2𝑟 in-
dicate that the random correlation in the training set is tolerable. This

5
H. Hadni and M. Elhallaouia Heliyon 8 (2022) e11537

Table 4. Statistical data from the external validation an electronegative substitution can improve the activity. This is consis-
of CoMSIA/EHA model. tent with the fact that all compounds with electronegative substituents
Parameter Validation Criteria CoMSIA/EHA at the R2 position show higher activity. Therefore, the presence of elec-
𝑄2 𝑄2 > 0.5 0.571 tronegative groups in R2 substitutions could have better activity.
r2 r2 > 0.6 0.715
| 2 | | 2 |
In the CoMSIA hydrophobic contour map (Fig. 4(b)), we observed
|𝑟0 − 𝑟′02 | |𝑟0 − 𝑟′02 | < 0.3 0.02
| | | | a gray contour near the pyrrolidine and piperidine of the R2 substitu-
k 0.85 < k < 1.15 1.026
tion, indicating that hydrophilic substitution is required in this region.
𝑟2 −𝑟20 𝑟2 −𝑟20
𝑟2 𝑟2
< 0.1 0.02 This is consistent with the fact that compound 1-6 with tert-butyl for-
K’ 0.85 < k′ < 1.15 0.973 mate as the hydrophilic group has a higher activity than compound
𝑟2 −𝑟0′ 2 𝑟2 −𝑟20
𝑟2 𝑟2
< 0.1 0.084 6-12, respectively. Thus, the presence of a hydrophilic group bound to
𝑟2𝑚 𝑟2𝑚 > 0.5 0.628 pyrrolidines or piperidines can increase the biological activity.
𝑟′𝑚2 𝑟′𝑚2 > 0.5 0.581 In the hydrogen bond acceptor contour map (Fig. 4(c)), the red
contours near the R2 substitution indicate that the substitution of the
hydrogen bond acceptor in this position is unfavourable. Thus, the
presence of hydrogen bond acceptor groups decreased the biological
activity, which is due to the nature of the receptor in this region which
can be a hydrogen bond acceptor. This can be explained by the fact
that compounds 24 and 28 with hydrogen bond donor groups show
better activities, and with the fact that compounds with hydrogen sub-
stitution in the R1 group are more active than compounds with fluorine
substitution in the same position. However, a small magenta contour
near the nitrogen atom of the pyrrolidine or piperidine substitution R2
is also observed indicating that the hydrogen bond acceptor atoms are
favourable in this position only. This can be explained by the fact that
compound 21 shows the best activities of the series. Generally, hydro-
gen bond acceptor groups are unfavorable to inhibitory activity. From
Table 2, we notice that the hydrogen bond acceptor field plays a key
role in predicting anticancer activity. In the case of the CoMSIA/EHA
model, the hydrogen bond acceptor field explains 63.6% of the vari-
ance, which explains why hydrogen bond acceptor groups are essential
for the inhibition of this process.

3.4. Molecular docking analysis

Fig. 3. The applicability domain of the CoMSIA/EHA model.
revealed that the results obtained from the original CoMSIA/EHA model
were not due to chance correlation. The results of the external valida-
tion test calculated by Eq. (3)-(8) with the Tropcha and Roy criteria for
the CoMSIA/EHA model are listed in Table 4.
The results in Table 4 reveal that the CoMSIA/EHA model is in per-
fect agreement with the Tropsha and Golbraikh as well as roy criteria.
The CoMSIA/EHA model passed all validation tests, showing a better
accuracy in predicting the activity of new compounds. Therefore, to de-
termine the applicability domain of this model, we used William’s graph
presented in Fig. 3.
The DA of the CoMSIA/EHA model was assessed by a leverage
analysis represented by a Williams diagram (Fig. 3). In the Williams
diagrams, the results indicate that all leverage values of the training
and test sets were below the critical leverage value (h* = 0.47), except
for one outlier of compound 29, which was above the critical leverage,
this compound belongs to the test set. The test set of the CoMSIA/EHA
model was accurately predicted because there were no outliers for the
training set. Therefore, we can reliably predict the anticancer activity
of new compounds using this model. Thus, CoMSIA/EHA contour maps
were used to analyse the structural requirements for the design of new
active compounds.
3.3. CoMSIA model visualization
The CoMSIA/EHA model was used to visualise the three-dimensional
equipotentiality map, using the compound 21 with the highest activity
as a template. Fig. 4 shows the electrostatic, hydrophobic and hydrogen
bond acceptor fields contour maps of the CoMSIA model.
In the CoMSIA electrostatic contour maps (Fig. 4(a)), we observed
a red contour near the pyrrolidine and piperidine of the R2 indicating
The molecular docking study was conducted to obtain information
on key structural requirements and to analyze the established interac-
tion with EGFRL858R/T790M/C797S protein. Fig. 5 presents the interaction
modes obtained by molecular docking for compounds 2 (IC50 = 0.58)
and 3 (IC50 = 0.012).
Molecular docking interaction of compound 3 with
EGFRL858R/T790M/C797S , as observed in Fig. 5 (a), showed three hy-
drogen bonding interactions with MET790, MET793 and SER797 at
distances of 3.07 Å, 1.65 Å and 2.52 Å, respectively. In addition, Pi-
sulfur interactions are observed between the two purine rings with
the amino acid MET792 (4.12 Å, 5.38 Å). However, compound 2 with
EGFRL858R/T790M/C797S (Fig. 5 (b)) forms only two hydrogen bonding
interactions with the amino acid with MET790 (2.19 Å) and MET793
(1.79 Å) and one halogen bonding interaction with the amino acid
ASP855 (2.69 Å), as well as Two Pi-sulfur interactions are observed
between the two purine rings with the amino acid MET790 (4.35 Å,
5.86 Å).
Compounds 3 and 2 formed two hydrogen bond and pi-sulfur inter-
actions with the same residues and position for both ligands. However,
the more active compound 3 formed an additional hydrogen bonding
interaction with the most important mutated residue, Ser797, in the
EGFR binding region, which enhanced the inhibitory activity. Interest-
ingly, the presence of these hydrogen bonds strengthened the binding
of the compounds to the protein and allowed the compounds to have
strong inhibitory activity.
According to Table 1, the fluorine substituent in the R1 group de-
creased the inhibitory activity compared to the hydrogen substituent,
this could be due to the formation of a halogen interaction with the fluo-
rine substituent that blocks the ligand in the active site, thus decreasing
the stability of the ligand-receptor complex. Furthermore, the CoM-
SIA/EHA hydrogen bond acceptor field contour map shows that ligands
with hydrogen bond acceptor groups in are unfavourable for anticancer

6
H. Hadni and M. Elhallaouia
Fig. 4. (a) Electrostatic, (b) hydrophobic and (c) Hydrogen bond acceptor for CoMSIA contour map analysis of compound 21.
activity. This observation is fully consistent with the 3D docking re-
sults which clearly show that the EGFRL858R/T790M/C797S protein is a
hydrogen bond acceptor at this position, validating our hypothesis in
the CoMSIA contour map section.
3.4.1. Molecular docking validation
In order to validate the ability of docking algorithms to pre-
dict the conformation of the ligand (Osimertinib) bound to the
EGFRL858R/T790M/C797S proteins, a self-docking of the crystal ligands
was performed to test the accuracy of the docking procedure. Fig. 6
shows the superimposed view between the conformation of the docked
ligand and native ligand, with an RMSD value of 0.869 Å less than 2 Å.
To visualise the quality of the docking poses in the protein, a visual in-
spection of the interactions between the crystallized and docked ligand
was performed (Fig. 7).
The results of the visual inspection show that we obtained the same
interaction modes as in the case of the experimental interaction of the
drug Osimertinib, observed in Fig. 7(a, b), this indicates a high reliabil-
ity of the docking protocols to produce the binding mode of the novel
EGFRL858R/T790M/C797S inhibitors.
3.5. Drug molecular design
The main objective of this study is to design new EGFR protein
inhibitors to overcome the L858R/T790M/C797S mutations. In this re-
gard, new drug candidates are designed using recommendations from
3D-QSAR and molecular docking analysis (Fig. 8) on the structural char-
acteristics of compound 21. In this study, ten 9H-purine derivatives
(N1 -N10 ) were designed to improve anticancer activity, the proposed
substitutions were taken from the ZINC database. Thus, these newly de-
signed compounds were aligned using compound 21 as a template. The
previously established CoMSIA/EHA model and molecular docking with
EGFRL858R/T790M/C797S protein were used to predict the activity of these
7
Heliyon 8 (2022) e11537
new compounds. The structure of the newly designed molecules, the
predicted pIC50 and IC50 values, the calculation of the leverage thresh-
old h* as well as the molecular docking interactions are presented in
Table 5.
The first step of a reliable prediction is to verify that all pre-
dicted new molecules belong to the application domain of the proposed
model. The CoMSIA/EHA model has a critical leverage value h* =
0.47, all the designed novel compounds have a threshold leverage value
h* lower than the critical leverage value, except for compounds N1
and N10 which could belong to another chemical family, this result
shows that N2-N9 compounds were reliably predicted anticancer ac-
tivity. Using the CoMSIA/EHA model, the compound T2 also showed
better activity than all compounds in the data set. Furthermore, we car-
ried out molecular docking for ten new compounds designed with the
EGFRL858R/T790M/C797S protein. The results indicate that almost all com-
pounds formed hydrogen bonding interactions with the amino acids at
positions 797 and 790 of the protein, both of which play an impor-
tant role in the therapeutic failure of the anticancer drug. However, the
inhibitory activity of compounds N1 and N10 cannot be reliably pre-
dicted, but the molecular docking results are encouraging and may lead
to a new inhibitor family. Overall, all compounds show a good level of
inhibitory activity and can theoretically overcome the problem of drug
resistance in lung cancer.
3.6. ADMET prediction
To verify that the designed compounds can become drugs, we use
the ADMET pharmacokinetic parameters by the online tool pkCSM [44].
The ADMET parameters of the newly designed compounds are listed in
Table 6.
An absorption value of less than 30% is considered poor intesti-
nal absorption, the ten compounds designed showed a value between
(67.671% and 97.823%) indicating good intestinal absorption, volume
H. Hadni and M. Elhallaouia
Fig. 5. 3D and 2D presentation of the interactions of compounds 3 and 2 in the binding sites of EGFRL858R/T790M/C797S protein. (a) Compound 3 (binding energy
−9.52 kcal/mol). (b) Compound 2: (binding energy −9.96 kcal/mol).
Fig. 6. Re-docking pose with RMSD value of 0.869 Å (Original = Green, Docked
= blue).
of distribution (VDss) is considered low if logVDss < −0.15 and high
if logVDss > 0.45, central nervous system (CNS) and blood-brain bar-
rier (BBB) permeability standard values (>−2 to <−3 LogPS and >0.3
to <−1 Log BB), respectively, for a given compound a LogBB < −1 cor-
8
Heliyon 8 (2022) e11537
responds to poor distribution to the brain, while LogBB >0.3 are likely
to cross the BBB and LogPS >3, to cross the BBB and LogPS > −2 con-
sidered to penetrate the CNS, while LogPS < −3 are difficult to move
into the CNS [45]. Thus, the compounds N1, N2, N6 and N8-N10 have
an excellent potential to cross the barriers.
The enzymatic metabolism refers to the chemical biotransforma-
tion of drugs in the human body, which plays a crucial role in the
metabolic stability of drugs in the body [46]. The cytochrome P450
enzymes (CYP1A2, CYP3A4, CYP2C19, CYP2D6 and CYP2C9) found in
the liver are the main enzymes of drug metabolism, being responsible
for the biotransformation of more than 90% of drugs. Inhibition of these
metabolising enzymes can increase the concentration of the active drug
in the body. In this study, CYP3A4 was the main human enzyme respon-
sible for the metabolism of the third-generation drug for treating NSCLC
[47, 48, 49]. The results show that all newly designed compounds ap-
pear to be CYP3A4 inhibitors, but only compounds N2, N5 and N6
appear to be CYP3A4 substrates. All newly designed compounds showed
a low total clearance value, which means accumulation and persistence
of the drugs in the body. Finally, compounds N1, N2, N6 and N8-
N10 showed negative toxicity. Overall, the newly designed compounds
H. Hadni and M. Elhallaouia
Fig. 7. (a) 2D visualization showing the interactions of the docked ligand. (b) 2D visualization showing the interactions of the native ligand.
Fig. 8. Summary of Structural requirements based on CoMSIA/EHA contour
map and molecular docking analyses.
N1, N2, N6 and N8-N10 exhibit good pharmacokinetic properties. The
results of this study may represent excellent drug candidates for the
treatment of NSCLC to overcome the L858R/T790M/C797S mutations
in EGFR-TKIs. Fig. 9 shows the 2D visualization of the molecular dock-
ing results of the best predicted compounds, which were then subjected
to a molecular dynamics study
3.7. Molecular dynamic simulation results
After performing molecular docking studies and ADMET properties
of the predicted compounds, MD simulation of the best candidates N1,
N2, N6 and N8-N10 was carried out, with RMSD and RMSF parameters
to analyse the stability of the target protein. The RMSD and RMSF plots
of the EGFRL858R/T790M/C797S complex by Osimertinib and with the best
predicted compounds are shown in Fig. 10 (a) and (b), respectively.
According to the RMSD plot analysis, all systems experienced a rapid
increase in RMSD values from 0.53 Å to 1.2 Å within 40 ns, which is
largely due to the initial kinetic shock experienced by all systems in
the molecular dynamics studies. Subsequently, the RMSD values of all
9
Heliyon 8 (2022) e11537
systems fluctuated within a similar distance range of 1.2 Å and 1.5
Å, indicating that all systems reached stability and equilibrium. The
most stable complexation was the experimental crystallographic struc-
ture of the EGFRL858R/T790M/C797S protein in combination with the drug
Osimertinib (6lud) obtained from the RCSB protein database with an
RMSD value of 1.082 Å. Osimertinib serves as the reference ligand in
this study. From the RMSD plot, it can be seen that all the proposed
ligands have a similar trajectory to the reference ligand with a slight
difference, the RMSD values of ligands N1, N2, N6, N8, N9 and N10
complexed with EGFRL858R/T790M/C797S were 1.231 Å, 1.137 Å, 1.173
Å, 1.169 Å, 1.159 Å and 1.251 Å, respectively. All ligands had a value
of about 1.3 Å and the N2 ligand had the best stability. A study by
Beura et al. [50] showed that an RMSD value of less than 3 Å is an
indicator of the conformational stability of protein-ligand complexes.
Therefore, the docking results of all predicted ligand complexations
with the EGFRL858R/T790M/C797S show higher conformational stability
of ligand-receptor complexes.
The RMSF trajectories assess the effect of ligand binding on the
flexibility of the protein during the MD simulation, which shows the
stability of the protein per residue during the simulation. The higher
the RMSF value, the more flexible the residue, while the lower the
RMSF value, the more stable the residue. In general, the majority of the
residuals have the same RMSF values, with higher fluctuations in the
different ranges, such as LYS754 (1.35 Å), LEU782 (1.14 Å), GLY874
(1.20 Å), SER921 (1.12 Å) and GLU1005 (1.11 Å), these residues
are not involved as they are located in the inactive regions of the
EGFRL858R/T790M/C797S protein. However, crucial residues in the active
site such as LEU718, ASP800, GLU804, MET790, MET793, SER797 and
ARG858 show smaller RMSF fluctuations of less than 0.4 Å, which could
be related to the generation of more hydrogen bonding interactions for
greater stability of the ligands with the EGFRL858R/T790M/C797S protein.
These results confirm the RMSD analysis that all predicted candidates
with the EGFRL858R/T790M/C797S protein show greater conformational
stability.
4. Conclusion
In summary, 3D-QSAR, docking, ADMET and MD simulation meth-
ods were performed for the design of new drug candidates capable
H. Hadni and M. Elhallaouia Heliyon 8 (2022) e11537

Table 5. The leverage threshold h* and predicted IC50 based on CoMSIA/EHA model and Molecular Docking interactions for the newly designed compounds.
Compounds Structures of newly designed molecules PIC50 IC50 leverage Interactions with EGFRL858R/T790M/C797S protein
(pred) (pred) threshold
Binding Number of Amino acid residues with
h*
affinity Hydrogen hydrogen bonding
(kcal/mol) bonds (distances)

MET793
N1 7.5432 0.0286 0.8065 −10.3 2
(1.81 Å, 2.54 Å)

SER797
(2.35 Å, 2.73 Å)
MET793
N2 9.1590 0.0007 0.3717 −10.55 4
(1.71 Å)
MET790
(2.57 Å)

ASP800
(2.92 Å)
SER797
(2.27 Å)
N3 8.8819 0.0013 0.3486 −9.17 4
MET793
(1.81 Å)
LEU718
(2.19 Å)

GLU804
(2.31 Å)
N4 7.0516 0.0887 0.4616 −9.95 2
SER797
(2.41 Å)

SER797
(2.22Å)
N5 8.6883 0.002 0.4160 −9.3 3
MET793
(2.04 Å, 2.71 Å)

10
H. Hadni and M. Elhallaouia Heliyon 8 (2022) e11537

Table 5 (continued)
Compounds Structures of newly designed molecules PIC50 IC50 leverage Interactions with EGFRL858R/T790M/C797S protein
(pred) (pred) threshold
Binding Number of Amino acid residues with
h*
affinity Hydrogen hydrogen bonding
(kcal/mol) bonds (distances)

MET790
(2.73 Å)
N6 7.0456 0.09 0.2360 −10.69 2
MET793
(1.78 Å)

SER797
(2.35 Å)
N7 7.2338 0.058 0.4589 −10.57 2
Met793
(1.91 Å)

SER797
(2.38Å)
N8 8.0158 0.0096 0.1332 −9.3 3
MET793
(2.05 Å, 2.53 Å)

SER797
(1.96 Å)
N9 7.5243 0.0299 0.2335 −9.2 3
MET793
(2.31 Å, 2.49 Å)

SER797
(2.24 Å, 2.90 Å)
MET793
N10 7.3045 0.0496 0.6577 −10.38 5
(2.21 Å, 2.16 Å)
MET790
(2.37 Å)

11
H. Hadni and M. Elhallaouia Heliyon 8 (2022) e11537

Fig. 9. 2D interactions of the best ligands N1, N2, N6 and N8-N10 in the binding sites of EGFRL858R/T790M/C797S protein.

12
H. Hadni and M. Elhallaouia Heliyon 8 (2022) e11537

Table 6. In silico ADMET prediction of newly designed inhibitors.

Compounds Absorption Distribution Metabolism Excretion Toxicity
Intestinal VDss BBB CNS Substrate Inhib itor Total AMES
absorption (human) permeability permeability CYP Clearance toxicity
(human) 2D6 3A4 1A2 2C19 2C9 2D6 3A4
Numeric Numeric Numeric Numeric Categorical (Yes/No) Numeric (Log Categorical
(% Absorbed) (Log L/kg) (Log BB) (Log PS) ml/min/kg) (Yes/No)
N1 90.063 0.556 −1.608 −2.823 No No No Yes Yes No Yes 0.758 No
N2 87.61 0.623 −1.43 −2.525 No Yes No Yes Yes No Yes 0.229 No
N3 67.671 0.432 −1.492 −3.755 No No No No Yes No Yes 0.485 Yes
N4 78.808 0.585 −1.709 −2.826 No No No Yes Yes No Yes 0.221 Yes
N5 82.792 −0.197 −1.252 −2.146 No Yes Yes Yes Yes No Yes 0.302 Yes
N6 97.823 −0.039 −1.485 −3.391 No Yes No Yes Yes No Yes 0.638 No
N7 78.764 0.036 −1.775 −3.944 No No No No Yes No Yes 0.149 Yes
N8 81.253 0.058 −1.917 −4.339 No No No No Yes No Yes 0.089 No
N9 90.08 0.063 −1.697 −4.004 No No No No Yes No Yes 0.328 No
N10 94.626 0.079 −1.489 −3.844 No No No Yes Yes No Yes 0.5 No

Fig. 10. a) The RMSD values of the EGFRL858R/T790M/C797S protein in complex with Osimertinib and six best ligands at 100 ns, and b) The RMSF values of the
EGFRL858R/T790M/C797S protein residues in complex with Osimertinib and six best.

of overcoming drug resistance in third generation NSCLC. In the 3D- Funding statement
QSAR study, the best selected model (CoMSIA/EHA) has high stability
and predictive ability, which were assessed using external validation, This research did not receive any specific grant from funding agen-
Y-randomisation test and applicability domain. The CoMSIA/SEHA con- cies in the public, commercial, or not-for-profit sectors.
tour map analyses provided a better understanding of the relationship
between structure and activity, which was a progression in guiding the Data availability statement
design of new potent compounds. The molecular docking analysis shows
the importance of hydrogen bonds established with key residues, which Data will be made available on request.
also confirmed the importance of residues such as MET790, MET793
and SER797 for the active site of EGFR protein. Based on the precise Declaration of interests statement
recommendation provided by 3D-QSAR and molecular docking analy-
sis, 10 new compounds with considerable activity were designed using
The authors declare no conflict of interest.
the virtual zinc base. The ADMET properties were used to select the best
pharmacokinetic profile of the proposed compounds. Finally, MD simu-
Additional information
lations verified the accuracy of the molecular docking results in terms of
reliability and stability, in which essential residues formed hydrogens
No additional information is available for this paper.
between the proposed compounds and the EGFR protein. The newly
designed compounds N1, N2, N6 and N8-N10 could be good drug can-
didates to overcome resistance to third generation drugs against NSCLC. References

[1] H. Sung, J. Ferlay, R.L. Siegel, M. Laversanne, I. Soerjomataram, A. Jemal, F. Bray,

Declarations
Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality
worldwide for 36 cancers in 185 countries, CA Cancer J. Clin. 71 (2021) 209–249.
Author contribution statement [2] Cancer, (n.d.), https://www.who.int/news-room/fact-sheets/detail/cancer. (Ac-
cessed 27 March 2021).
Hanine Hadni: Conceived and designed the experiments; Analyzed [3] K. Seegobin, U. Majeed, N. Wiest, R. Manochakian, Y. Lou, Y. Zhao, Immunotherapy
in non-small cell lung cancer with actionable mutations other than EGFR, Front.
and interpreted the data; Wrote the paper. Oncol. 11 (2021) 5040.
Menana Elhallaoui: Contributed reagents, materials, analysis tools [4] F. Ciardiello, G. Tortora, EGFR antagonists in cancer treatment, N. Engl. J. Med. 358
or data; Wrote the paper. (2008) 1160–1174.

13
H. Hadni and M. Elhallaouia
[5] A. Harandi, A.S. Zaidi, A.M. Stocker, D.A. Laber, Clinical efficacy and toxicity of
anti-EGFR therapy in common cancers, J. Oncol. (2009).
[6] S.V. Sharma, D.W. Bell, J. Settleman, D.A. Haber, Epidermal growth factor receptor
mutations in lung cancer, Nat. Rev. Cancer 7 (2007) 169–181.
[7] I. Solassol, F. Pinguet, X. Quantin, FDA- and EMA-approved tyrosine kinase in-
hibitors in advanced EGFR-mutated non-small cell lung cancer: safety, tolerability,
plasma concentration monitoring, and management, Biomolecules 9 (2019) 668.
[8] M.H. Cohen, G.A. Williams, R. Sridhara, G. Chen, W.D. McGuinn, D. Morse, S. Abra-
ham, A. Rahman, C. Liang, R. Lostritto, A. Baird, R. Pazdur, United States food and
drug administration drug approval summary: gefitinib (ZD1839; Iressa) tablets, Clin.
Cancer Res. 10 (2004) 1212–1218.
[9] A.F. Gazdar, Activating and resistance mutations of EGFR in non-small-cell lung can-
cer: role in clinical response to EGFR tyrosine kinase inhibitors, Oncogene 28 (Suppl
1) (2009) S24–S31.
[10] M. Tiseo, M. Bartolotti, F. Gelsomino, P. Bordi, Emerging role of gefitinib in the
treatment of non-small-cell lung cancer (NSCLC), Drug Des. Devel. Ther. 4 (2010)
98.
[11] W. Pao, V. Miller, M. Zakowski, J. Doherty, K. Politi, I. Sarkaria, B. Singh, R. Hee-
lan, V. Rusch, L. Fulton, E. Mardis, D. Kupfer, R. Wilson, M. Kris, H. Varmus, EGF
receptor gene mutations are common in lung cancers from “never smokers” and are
associated with sensitivity of tumors to gefitinib and erlotinib, Proc. Natl. Acad. Sci.
101 (2004) 13306–13311.
[12] M. Singh, H.R. Jadhav, Targeting non-small cell lung cancer with small-molecule
EGFR tyrosine kinase inhibitors, Drug Discov. Today 23 (2018) 745–753.
[13] D. Westover, J. Zugazagoitia, B.C. Cho, C.M. Lovly, L. Paz-Ares, Mechanisms of
acquired resistance to first- and second-generation EGFR tyrosine kinase inhibitors,
Ann. Oncol. 29 (2018) i10–i19.
[14] M.R.V. Finlay, M. Anderton, S. Ashton, P. Ballard, P.A. Bethel, M.R. Box, R.H. Brad-
bury, S.J. Brown, S. Butterworth, A. Campbell, C. Chorley, N. Colclough, D.A.E.
Cross, G.S. Currie, M. Grist, L. Hassall, G.B. Hill, D. James, M. James, P. Kemmitt,
T. Klinowska, G. Lamont, S.G. Lamont, N. Martin, H.L. McFarland, M.J. Mellor, J.P.
Orme, D. Perkins, P. Perkins, G. Richmond, P. Smith, R.A. Ward, M.J. Waring, D.
Whittaker, S. Wells, G.L. Wrigley, Discovery of a potent and selective EGFR inhibitor
(AZD9291) of both sensitizing and T790M resistance mutations that spares the wild
type form of the receptor, J. Med. Chem. 57 (2014) 8249–8267.
[15] J.J. Chabon, A.D. Simmons, A.F. Lovejoy, M.S. Esfahani, A.M. Newman, H.J. Har-
ingsma, D.M. Kurtz, H. Stehr, F. Scherer, C.A. Karlovich, T.C. Harding, K.A. Durkin,
G.A. Otterson, W.T. Purcell, D.R. Camidge, J.W. Goldman, L.V. Sequist, Z. Pi-
otrowska, H.A. Wakelee, J.W. Neal, A.A. Alizadeh, M. Diehn, Circulating tumour
DNA profiling reveals heterogeneity of EGFR inhibitor resistance mechanisms in
lung cancer patients, Nat. Commun. 7 (2016).
[16] H. Hadni, M. Bakhouch, M. Elhallaoui, 3D-QSAR, molecular docking, DFT and AD-
MET studies on quinazoline derivatives to explore novel DHFR inhibitors, J. Biomol.
Struct. Dyn. (2021) 1–15.
[17] S. Sarvagalla, S.B. Syed, M.S. Coumar, An overview of computational methods, tools,
servers, and databases for drug repurposing, in: Silico Drug Des., Elsevier, 2019,
pp. 743–780.
[18] H. Hadni, M. Elhallaoui, 3D-QSAR, docking and ADMET properties of aurone ana-
logues as antimalarial agents, Heliyon 6 (2020) e03580.
[19] H. Lei, S. Fan, H. Zhang, Y.J. Liu, Y.Y. Hei, J.J. Zhang, A.Q. Zheng, M.
Xin, S.Q. Zhang, Discovery of novel 9-heterocyclyl substituted 9H-purines as
L858R/T790M/C797S mutant EGFR tyrosine kinase inhibitors, Eur. J. Med. Chem.
186 (2020) 111888.
[20] G. Klebe, U. Abraham, T. Mietzner, Molecular similarity indices in a comparative
analysis (CoMSIA) of drug molecules to correlate and predict their biological activ-
ity, J. Med. Chem. 37 (1994) 4130–4146.
[21] R.R. Mittal, L. Harris, R.A. McKinnon, M.J. Sorich, Partial charge calculation method
affects CoMFA QSAR prediction accuracy, J. Chem. Inf. Model. 49 (2009) 704–709.
[22] M.J.D. Powell, Restart procedures for the conjugate gradient method, Math. Pro-
gram. 12 (1977) 241–254.
[23] S. Wold, A. Ruhe, H. Wold, W.J. Dunn III, The collinearity problem in linear regres-
sion. The partial least squares (PLS) approach to generalized inverses, SIAM J. Sci.
Stat. Comput. 5 (1984) 735–743.
[24] K. Roy, On some aspects of validation of predictive quantitative structure-activity
relationship models, Expert Opin. Drug Discov. 2 (2007) 1567–1577.
[25] K. Roy, I. Mitra, On various metrics used for validation of predictive QSAR models
with applications in virtual screening and focused library design, Comb. Chem. High
Throughput Screen. 14 (2011) 450–474.
14
Heliyon 8 (2022) e11537
[26] A. Golbraikh, A. Tropsha, Beware of q2!, J. Mol. Graph. Model. 20 (2002) 269–276.
[27] K. Roy, S. Kar, P. Ambure, On a simple approach for determining applicability do-
main of QSAR models, Chemom. Intell. Lab. Syst. 145 (2015) 22–29.
[28] T.I. Netzeva, A.P. Worth, T. Aldenberg, R. Benigni, M.T.D. Cronin, P. Gramatica, J.S.
Jaworska, S. Kahn, G. Klopman, C.A. Marchant, G. Myatt, N. Nikolova-Jeliazkova,
G.Y. Patlewicz, R. Perkins, D.W. Roberts, T.W. Schultz, D.T. Stanton, J.J.M. van
de Sandt, W. Tong, G. Veith, C. Yang, Current status of methods for defining the
applicability domain of (quantitative) structure-activity relationships, Altern. Lab.
Anim. 33 (2005) 155–173.
[29] S. Kar, K. Roy, J. Leszczynski, Applicability domain: a step toward confident predic-
tions and decidability for QSAR modeling, Methods Mol. Biol. 1800 (2018) 141–169.
[30] K. Kashima, H. Kawauchi, H. Tanimura, Y. Tachibana, T. Chiba, T.
Torizawa, H. Sakamoto, CH7233163 overcomes osimertinib-resistant EGFR-
Del19/T790M/C797S mutation, Mol. Cancer Ther. 19 (2020) 2288–2297.
[31] D.S. BIOvIA, Discovery studio modeling environment, San Diego, Dassault Syst. Re-
lease, 4, 2015.
[32] G.M. Morris, R. Huey, W. Lindstrom, M.F. Sanner, R.K. Belew, D.S. Goodsell, A.J.
Olson, AutoDock4 and AutoDockTools4: automated docking with selective receptor
flexibility, J. Comput. Chem. 30 (2009) 2785–2791.
[33] G.M. Morris, D.S. Goodsell, R.S. Halliday, R. Huey, W.E. Hart, R.K. Belew, A.J.
Olson, AutoDock-related material automated docking using a Lamarckian genetic al-
gorithm and an empirical binding free energy function, J. Comput. Chem. 19 (1998)
1639–1662.
[34] K. Onodera, K. Satou, H. Hirota, Evaluations of molecular docking programs for
virtual screening, J. Chem. Inf. Model. 47 (2007) 1609–1618.
[35] G.L. Warren, C.W. Andrews, A.-M. Capelli, B. Clarke, J. LaLonde, M.H. Lambert,
M. Lindvall, N. Nevins, S.F. Semus, S. Senger, G. Tedesco, I.D. Wall, J.M. Woolven,
C.E. Peishoff, M.S. Head, A critical assessment of docking programs and scoring
functions, J. Med. Chem. 49 (2006) 5912–5931.
[36] Hanine Hadni, Menana Elhallaoui, 2D and 3D-QSAR, molecular docking and AD-
MET properties in silico studies of azaaurones as antimalarial agents, New J. Chem.
(2020).
[37] L.L.G. Ferreira, A.D. Andricopulo, ADMET modeling approaches in drug discovery,
Drug Discov. Today 24 (2019) 1157–1165.
[38] C.Y. Jia, J.Y. Li, G.F. Hao, G.F. Yang, A drug-likeness toolbox facilitates ADMET
study in drug discovery, Drug Discov. Today 25 (2020) 248–258.
[39] J.C. Phillips, R. Braun, W. Wang, J. Gumbart, E. Tajkhorshid, E. Villa, C. Chipot, R.D.
Skeel, L. Kalé, K. Schulten, Scalable molecular dynamics with NAMD, J. Comput.
Chem. 26 (2005) 1781–1802.
[40] S. Jo, T. Kim, V.G. Iyer, W. Im, CHARMM-GUI: a web-based graphical user interface
for CHARMM, J. Comput. Chem. 29 (2008) 1859–1865.
[41] W. Im, S. Seefeld, B. Roux, A grand canonical Monte Carlo–Brownian dynamics
algorithm for simulating ion channels, Biophys. J. 79 (2000) 788–801.
[42] H. Hadni, A. Fitri, A.T. Benjelloun, M. Benzakour, M. Mcharfi, Evaluation of
flavonoids as potential inhibitors of the SARS-CoV-2 main protease and spike RBD:
molecular docking, ADMET evaluation and molecular dynamics simulations, J. In-
dian Chem. Soc. 99 (2022) 100697.
[43] W. Humphrey, A. Dalke, K. Schulten, VMD: visual molecular dynamics, J. Mol.
Graph. 14 (1996) 33–38.
[44] D.E.V. Pires, T.L. Blundell, D.B. Ascher, pkCSM: predicting small-molecule pharma-
cokinetic and toxicity properties using graph-based signatures, J. Med. Chem. 58
(2015) 4066–4072.
[45] D.E. Clark, In silico prediction of blood–brain barrier permeation, Drug Discov. To-
day 8 (2003) 927–933.
[46] S. Kok-Yong, L. Lawrence, Drug distribution and drug elimination, in: Basic Phar-
macokinet. Concepts Some Clin. Appl., InTech, 2015.
[47] D.R. Duckett, M.D. Cameron, Metabolism considerations for kinase inhibitors in can-
cer treatment, Expert Opin. Drug Metab. Toxicol. 6 (2010) 1193.
[48] M.K. Bollinger, A.S. Agnew, G.P. Mascara, Osimertinib: a third-generation tyrosine
kinase inhibitor for treatment of epidermal growth factor receptor-mutated non-
small cell lung cancer with the acquired Thr790Met mutation, J. Oncol. Pharm.
Pract. 24 (2018) 379–388.
[49] A. Kenneth MacLeod, D. Lin, J.T.J. Huang, L.A. McLaughlin, C.J. Henderson, C.
Roland Wolf, Identification of novel pathways of osimertinib disposition and poten-
tial implications for the outcome of lung cancer therapy, Clin. Cancer Res. 24 (2018)
2138–2147.
[50] S. Beura, P. Chetti, In-silico strategies for probing chloroquine based inhibitors
against SARS-CoV-2, J. Biomol. Struct. Dyn. 39 (2021) 3747–3759.