5.2
Impact Factor
Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
Search in posts
Search in pages
Filter by Categories
Corrigendum
Current Issue
Editorial
Erratum
Full Length Article
Full lenth article
Letter to Editor
Original Article
Research article
Retraction notice
Review
Review Article
SPECIAL ISSUE: ENVIRONMENTAL CHEMISTRY
5.3
Impact Factor
Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
Search in posts
Search in pages
Filter by Categories
Corrigendum
Current Issue
Editorial
Erratum
Full Length Article
Full lenth article
Letter to Editor
Original Article
Research article
Retraction notice
Review
Review Article
SPECIAL ISSUE: ENVIRONMENTAL CHEMISTRY
View/Download PDF

Translate this page into:

Original Article
:19;
7592025
doi:
10.25259/AJC_759_2025

Design and in silico evaluation of quinazoline derivatives for hepatic cancer: QSAR modeling, ADMET profiling, molecular docking, and dynamics simulations

 Department of Chemistry, Turabah University College, Taif University, Taif, Saudi Arabia

*Corresponding author: E-mail address: mtotaibi@tu.edu.sa (M. Alotaibi)

Received: , Accepted: ,
© 2026 Arabian Journal of Chemistry - Published by Scientific Scholar
Licence
This is an open-access article distributed under the terms of the Creative Commons Attribution-Non Commercial-Share Alike 4.0 License, which allows others to remix, transform, and build upon the work non-commercially, as long as the author is credited and the new creations are licensed under the identical terms.

Abstract

This study focuses on the development of a quantitative structure-activity relationship (QSAR) model to predict the anticancer activity of quinazoline-based compounds, specifically targeting hepatic cancer (HepG2 cells). A dataset of 35 quinazoline derivatives with known anticancer activity was sourced from the literature. The compound structures were constructed using ChemDraw, and energy minimization was performed using the MM2 force field in Chem3D Pro. Molecular descriptors were calculated (1800 descriptors) using PaDEL Descriptor software, and QSAR modeling was conducted using QSARINS software. The dataset, including IC50 values, was divided into a training set (80%) and a test set (20%). Two QSAR models were developed: Model 1 demonstrated a high performance with an R2 value of 0.9671 and an adjusted R2 of 0.9600, while Model 2, selected as the optimal model, exhibited a stronger correlation (R2 = 0.9763, adjusted R2 = 0.9706) and minimal overfitting (Friedman’s lack of fit = 0.0185). Both models were validated through internal and external tests, including Leave-One-Out (LOO) and Leave-Many-Out (LMO) cross-validation, Y-scrambling, and applicability domain analysis. Model 2 showed excellent predictive accuracy, with key descriptors such as ATSC4m, AATSC7i, and GATS5c significantly influencing the biological activity predictions. Based on the best-performing QSAR model and SAR studies, 35 novel compounds (AA1-AA35) were designed, and their pIC50 values were predicted. The top 15 compounds were further evaluated through docking, ADMET (absorption, distribution, metabolism, excretion, and toxicity) analysis, and MD simulations. AA19 showed the best binding affinity (-8.30 kcal/mol), forming stable hydrogen bonds with ASP86, LEU83, and GLU12, and Pi-alkyl interactions with VAL18, ALA144, and LEU134. The complex remained stable, with protein RMSD between 1.75-2.0 Å and ligand RMSD starting at 1.2 Å. RMSF values remained below 2.4 Å, indicating minimal residue fluctuation.

Keywords

ADMET
Cancer
Molecular docking
Molecular dynamics
QSAR
Quinazoline

1. Introduction

Cancer is the world’s second most common cause of mortality, as indicated in recent reports from the World Health Organization [1]. In 2020 alone, there were an estimated 19 million new cancer cases globally, with nearly 10 million deaths attributed to the disease. According to GLOBOCAN 2022 estimates, liver cancer accounted for approximately 866,136 new cases globally, with 600,676 cases in males and 265,460 in females [2].

Cancer treatments typically involve a multifaceted approach, combining surgery, chemotherapy, radiation, targeted therapies, hormone treatments, immunotherapy, and bone marrow transplants [3]. Various combinations of anticancer drugs are often employed to induce early remission [4]. In cases where tumors measure under 5 cm, surgical removal can lead to a positive prognosis. However, because early symptoms are usually subtle or absent, many patients remain undiagnosed until the disease has reached an advanced stage, which contributes to high mortality rates. The development of liver cancers, particularly hepatocellular carcinoma, involves complex pathological mechanisms [5]. Managing this form of cancer often necessitates an integrative treatment strategy, where chemotherapy complements surgical interventions.

A combined therapeutic approach offers improved effectiveness over monotherapy, as it can target multiple key pathways, often resulting in additive or synergistic effects [4]. Beyond inhibiting tumor growth and reducing metastasis, this strategy can also mitigate toxicity and lessen the likelihood of drug resistance [6]. As a result, combination chemotherapy has become a widely adopted approach in clinical settings [7]. Despite these advantages, combination therapies can be costly, involve complex dosing regimens, and are associated with side effects and potential resistance [8]. This underscores the urgent need to develop new, more accessible, and safer treatments. Given these considerations, advancing a novel therapeutic approach for liver cancer remains a critical goal.

In recent years, a range of new bioactive compounds have been identified that specifically target molecular pathways crucial to the onset and progression of various cancers, enabling more focused treatment strategies [9]. Compounds containing heterocyclic structures have attracted considerable interest in medicinal chemistry [10]. Numerous synthesized heterocyclic molecules have demonstrated anti-cancer potential, promoting apoptosis across different cancer cell types. These compounds are tailored to interact with specific targets, disrupting critical biological pathways that contribute to cancer development [11].

Considerable efforts have focused on synthesizing a variety of quinazoline-based compounds, given their notable synthetic relevance and broad spectrum of therapeutic effects [12]. Quinazoline derivatives form a prominent class of synthetic compounds with substantial potential for developing anticancer drugs [13]. For instance, marketed drugs like nolatrexed and raltitrexed incorporate the quinazoline core structure (Figure 1a) [14]. Additionally, the quinazoline motif is central to several FDA-approved anticancer agents [15], including erlotinib, gefitinib, afatinib, and dacomitinib (Figure 1b), which are currently utilized in treating cancers such as colon, breast, prostate, and lung cancers [16].

(a) Quinazolinone derivatives with potent anticancer activity. (b) Drugs based on the quinazoline scaffold. (c) Quinazoline derivatives with strong anticancer properties. (d) Strategy adopted in the present study.
Figure 1.
(a) Quinazolinone derivatives with potent anticancer activity. (b) Drugs based on the quinazoline scaffold. (c) Quinazoline derivatives with strong anticancer properties. (d) Strategy adopted in the present study.

Additionally, certain quinazoline derivatives exhibit anticancer activity by targeting kinases involved in cell division regulation [17]. Cyclin-dependent kinases (CDKs), which belong to the serine-threonine kinase family, are essential in managing cell cycle progression [18]. Regulating CDK activity has emerged as a therapeutic strategy, as its dysregulation is intricately linked to abnormal cell division. CDK2 plays a pivotal role in DNA replication [19]. Quinazoline compounds have also been identified as inhibitors of both CDK4/D1 and CDK2/E, which are associated with the development of various cancer types [20]. These compounds can influence pathways involving unregulated cell growth, cell cycle disruptions, and genomic instability, contributing to cancer progression [21,22].

Sielecki and colleagues conducted an extensive study on the quinazoline scaffold, focusing on modifications at positions 2, 4, and 6. Their findings revealed that a quinazoline analog with an alkyl substitution at C-4 (Compound I, Figure 1c) was the most effective inhibitor of CDK2, exhibiting an IC50 of 0.6 μM [23]. In another investigation by Giulia Rodrigues Stringhetta, 2-arylquinazoline-chalcone hybrids have shown that substitution at the C2 position with an NH-linked chalcone derivative is critical for inducing cell cycle arrest and apoptosis via CDK2 and ATR inhibition in breast cancer cells. Compounds demonstrated potent cytotoxicity, highlighting the pharmacological relevance of the C2-NH linkage [24]. Bathini et al. explored modifications at the C-2 position by incorporating an aniline ring, which led to significant CDK4 inhibition in 2,7,8-trisubstituted quinazolines, achieving an IC50 of 0.007 μM (Compound II, Figure 1c) [25]. Given these insights, there remains a need for a selective CDK2 inhibitor with minimized off-target effects. Therefore, we aimed to design novel quinazoline analogs via computer tools specifically targeting CDK2 inhibition.

In this study, we focused on the development of a quantitative structure-activity relationship (QSAR) model to predict the anticancer activity of quinazoline-based compounds, specifically targeting hepatic cancer (HepG2 cells). This work builds on the findings of Gopalakrishnan Venkatesan’s team, who previously synthesized, spectroscopically characterized, and evaluated the biological activity of 35 quinazoline derivatives on multiple cancer cell lines [26]. For our investigation, we have focused on cytotoxicity data against liver cancer cells (HepG-2), utilizing IC50 values from 35 quinazoline derivatives (Figure 1d). We further emphasize the exploration of both ligand- and structure-based computational approaches to discover a new, effective scaffold against liver cancer cells (HepG-2). Subsequently, molecular docking, ADMET (absorption, distribution, metabolism, excretion, and toxicity) pharmacokinetic profiling, and molecular dynamics simulations were performed. Molecular docking analyses were applied to investigate the binding affinities and amino acid interactions. Erlotinib was selected as the standard reference compound due to its reported clinical efficacy in hepatocellular carcinoma and its documented off-target interactions with kinases such as CDK2, making it suitable for comparative docking analysis. Molecular Dynamics simulations were performed to predict the stability through parameters like root mean square fluctuation (RMSF), root mean square deviation (RMSD), and various ligand properties.

2. Materials and Methods

2.1. Quantity structure-activity relationship

The compound library was obtained from the reported dataset of quinazoline analogs as potent agents against hepatocarcinoma by Gopalakrishnan Venkatesan, et al. [26], and the structures of these compounds have been presented in Table 1. A dataset of 35 compounds, with an experimentally reported IC50 inhibition constant against the HepG2 cell line, was selected and compiled. These IC50 values, indicative of biological activity, were converted into pIC50 values using the formula pIC50 = -log (IC50). These pIC50 values will serve as the dependent variable in the analysis.

Table 1. Structural information of dataset compounds with diverse substitutions.
Compound Substitution Exp IC50(µM) HepG-2 Compound Substitution ExpIC50(µM) HepG-2
A1 H 15.8 A19 2,5-F 4.24
A2 4-Cl 5.22 A20 4-Cl, 3-NO2 12.48
A3 4-F 4.51 A21 4-F, 3-NO2 11.3
A4 4-Br 5.76 A22 4 Br, 2F 3.97
A5 4-I 6.92 A23 4-CH3, 3-Cl 3.49
A6 4-NO2 8.68 A24 4-CH3, 3-OH 3.08
A7 4-CH3 4.48 A25 4-OCH3, 3-Cl 2.25
A8 4-OCH3 2.42 A26 4-NO2, 2-OH 3.91
A9 4-(OCF2) 7.28 A27 4-OH, 5-Cl 2.05
A10 4-OH 2.65 A28 2-OH, 6-CH3 2.88
A11 4-Ethynyl 4.34 A29 3-Cl 14.64
A12 4-Tert butyl 5.97 A30 3-F 11.5
A13 4-CN 7.8 A31 3-Br 16.84
A14 4-SCH3 5.23 A32 3-I 18.2
A15 3,4-CH3 5.12 A33 3-ethynyl 7.42
A16 3,5-OCH3 2.55 A34 3Cl, 4-OCH2C6H4(3F) 0.102
A17 3,4,5-OCH3 2.69 A35 3Cl, 4-OCH2C6H4(3SO2CH3) 0.058
A18 2-Morpholine 1.5

PaDEL software was utilized to generate molecular for the dataset, yielding around 1800 descriptors, focusing on 1D and 2D descriptors. The 1D descriptors include simple molecular properties such as molecular weight and atom count, while the 2D descriptors comprise topological indices, connectivity, and molecular graph-based properties. These descriptors form the independent variables in the dataset [27].

The dataset containing both dependent and independent variables was used for QSAR analysis performed through QSARINS software 2.2.4. The efficiency of modelling and redundancy minimization of descriptors were performed. Descriptors that shared more than 80% of the same value as constant or nearly constant and descriptors showing high inter-correlation greater than 95% were removed as pretreatment. After applying these criteria, 480 descriptors were retained for the regression model development. The dataset was subsequently divided into 80% for model building (training set) and 20% for validation (test set) by a random selection method.

QSARINS software utilized genetic algorithms (GA) for model generation, applying ordinary least squares (OLS) to derive relationships between chemical structures and descriptors via multiple linear regression (MLR) analysis. Model generation was performed with specific parameters: a significance level of 0.05, fitness using Q2LOO, and the GA configured to explore up to five descriptors, with 2000 iterations, a population size of 10, and a mutation rate of 20%. The QUIK rule (Q Under the Influence of K) was employed to ensure high collinearity in the models. Based on the multi-criteria decision making (MCDM) approach, the best-scoring model, incorporating five descriptors, was selected for structural analysis in the QSAR framework.

According to the Organization for economic co-operation and development (OECD) principles, the reliability and validity of the QSAR model were analyzed. The model’s robustness and predictive capability were confirmed through both internal and external validation methods. Internal validation involved leave-one-out (LOO) and leave-many-out (LMO) cross-validation, while external validation metrics included a Y-randomization test. An assessment of the model’s applicability domain was conducted to evaluate the structural range to perform reliable predictions. A Williams plot, illustrating leverage values against standardized residuals, was employed to detect influential observations and potential outliers within the data.

2.2. Physico-chemical properties

The pharmacological effectiveness of a drug is strongly associated with its physicochemical properties, which are crucial in determining its pharmacokinetic behavior. Core properties such as the partition coefficient (Log P), molecular weight, hydrogen bond donors and acceptors, and topological polar surface area (TPSA) are critical for assessing a compound’s absorption, distribution, and bioavailability. These characteristics can be computed using online resources like Molinspiration software, which enables precise evaluation of molecular attributes. Lipinski’s Rule of 5 serves as a recommended guideline, with Log P ≤ 5, molecular weight ≤ 500, no more than 10 hydrogen bond acceptors (HBAs), and no more than five hydrogen bond donors (HBDs) are likely to exhibit favorable bioavailability. Deviations from these parameters can often suggest potential challenges with absorption and therapeutic efficacy [28].

2.3. ADME properties

An in silico ADME evaluation of compounds to provide pharmacokinetic characteristics was carried out using the SwissADME web tool. This early-stage assessment supports the prediction of key ADME parameters for novel chemical entities. By entering the compound’s SMILES notation, the tool generated predictions for lipophilicity (Log P), solubility, gastrointestinal (GI) absorption, and blood-brain barrier (BBB) permeability. These descriptors play a crucial role in assessing the compound’s potential for effective absorption and systemic distribution, aiding in the selection of promising drug-like candidates [29].

2.4. Toxicity

Progress in computational methodologies has substantially enhanced toxicity prediction capabilities, enabling essential safety assessments of chemical compounds through in silico methods. These approaches facilitate a comprehensive evaluation of various toxicity parameters, including hepatotoxicity, carcinogenicity, immunotoxicity, and mutagenicity. Using the ProTox-II server, toxicity profiles of newly designed compounds can be predicted and analyzed efficiently, contributing to safer drug development while reducing reliance on extensive in vitro or in vivo testing.

2.5. Molecular docking

Molecular docking was conducted using AutoDock 4.2 software. Initially, the selected ligand was imported into ChemDraw3D 16.0 via its .mol format, converted to a 3D structure, and subjected to energy minimization, resulting in a PDB file generation. The docking studies were performed against the CDK2 with the PDB ID 2A4L. Protein preparation involved removing non-essential residues and water molecules, followed by the addition of polar hydrogens and Kollman charges to ensure valency balance. The PDB files were then converted to PDBQT format by setting torsions and saving the modified structure.

A grid box was created by aligning the macromolecule’s dimensions along a 90-degree axis with dimensions of 126 points in x, y, and z axis with 0.375 spacing. The grid parameter file (.gpf) was saved. The docking process was set up by selecting search and genetic parameters, specifically the Lamarckian genetic algorithm, and generating dock parameter files (.dpf). These were processed into log files via the command prompt, with results saved in .dlg format. Post-docking analysis involved activating the ligand-protein conformations and examining their interactions. The .dlg files were further analyzed in BIOVIA Discovery Studio, allowing visualization of ligand-protein interactions in both 2D and 3D representations.

2.6. Molecular dynamic simulation

Molecular dynamics simulations were performed using Desmond V 5.9 provided by D.E. Shaw Research group (Schrödinger LLC) to analyze the dynamic behavior of docked protein-ligand complexes under physiological conditions. The protein-ligand complex was preprocessed using the Protein Preparation Wizard in Maestro, optimized at pH 7.0, and refined with default parameters. A 10 Å orthorhombic box was generated around the complex, solvated using the TIP3P water model, and neutralized with counter ions (0.15 M). Equilibration was conducted using NPT and NVT ensembles with Martyna-Tobias Klein and Nosé-Hoover methods at 300 K over 20,000 iterations. The production run spanned 50 nanoseconds under periodic boundary conditions. Simulated complexes were analyzed using RMSD, RMSF, and protein-ligand contacts [30-32].

3. Results and Discussion

3.1. Quantitative structure-activity relationship (QSAR)

3.1.1. Model information

The regression model 1 demonstrated strong performance with an R2 value of 0.9671 and an adjusted R2 of 0.9600, indicating that 96.71% of the variance in the dependent variable is explained by the model. The model showed a minimal difference (0.0071) between R2 and adj R2, suggesting low overfitting of the model. The Friedman’s lack of fit (LOF) was low at 0.0220, and the condition index (Kxx) was 0.2871, with a delta K of 0.1020, implying stable model parameters. The Concordance Correlation Coefficient (CCC tr) for the training set was high at 0.9833, the standard error of the regression (s) was 0.1092, and the F-statistic value was higher at 135.4196, exhibiting the model’s statistical significance. In terms of internal validation, the LOO cross-validation Q2loo score was 0.9473, with less difference between R2 and Q2loo (0.0199), confirming robust internal predictive power. The LMO Q2lmo was 0.9320, with a Y-scrambling test result showing R2Yscr of 0.1740 and Q2Yscr of -0.3640, confirming that the model’s predictions are not random and have a robust internal validation framework. The applicability domain analysis using the leverage plot identified three compounds, A1, A17, and A20, as outliers, with a leverage value of 0.621 and exceeding 2.5 standard deviations (Figure S1). These outliers impacted the model’s predictive accuracy and robustness. Consequently, these compounds were removed, and a second model was generated.

The QSAR model equation is as follows (Eqs. 1, 2).

Figure S1

(1)
Model   1 :   pIC 5 0 = − 3 . 1198 − 0.0 615   AATSC8m + 0. 9 0 32 GATS1i + 0.0 454  VE3 _ Dzv − 9 . 1428  hmin + 19 . 3273 ETA _ BetaP _ ns _ d

(2)
Model   2 : pIC 5 0 = 5 . 5725 − 0.000 2  ATSC4m − 1 .0 45 0 AATSC7i − 1 . 8843  GATS5c − 1 . 1 0 46 SpMax5 − Bhm − 9 .0 433  hmin

3.1.2. Characteristics of model 2

Model 2 was selected as the optimal model, displaying a strong correlation coefficient R2 of 0.9763 and an adjusted R2 of 0.9706. With an R2 value above 0.6, the model demonstrates excellent predictive capability. The model’s statistical parameters have been summarized in Tables 2 and 3 present the contribution of each descriptor and the predicted activity values. The difference between R2 and R2adj is 0.0056, and the standard error of estimate (s) is 0.0970, indicating minimal variance in the model’s predictions.

Table 2. Statistical parameters of the QSAR model 1 and 2.
Statistical parameters Threshold value Model 1 Model 2
R2 > 0.6 0.9671 0.9763
S < 0.3 0.1092 0.0970
R2adj > 0.6 0.9600 0.9706
R2 – R2adj < 0.3 0.0071 0.0056
LOF < 0.3 0.0220 0.0185
Kxx 0 ≤ Kxx ≤ 1 0.2871 0.3240
ΔK < 1 0.1020 0.0683
RMSE tr Better < 0.3 0.0973 0.0855
MAE tr Better < 0.3 0.0835 0.0718
CCC tr > 0.85 0.9833 0.9880
F Higher than the theoretical value 135.4196 172.9347
Q2loo (r2 cv) > 0.5 0.9473 0.9551
R2 – Q2 loo < 0.3 0.0199 0.0212
Q2 lmo > 0.5 0.9320 0.9463
PRESS cv Lower value is better 0.4400 0.3740
CCC cv > 0.85 0.9730 0.9775
RMSE cv < 0.3 0.1232 0.1177
MAE cv < 0.3 0.1081 0.0946
R2Yscr < R2 (smallest is better) 0.1740 0.1933
Q2Yscr < Q2 (smallest is better) -0.3640 -0.4139
R2ext > 0.6 0.4924 0.5011
RMSE ext Better < 0.3 0.2485 0.1616
MAE ext Better < 0.3 0.2205 0.1285
Q2-F1 > 0.7 0.5141 -0.0681
Q2-F2 > 0.7 0.3001 -0.4163
Q2-F3 > 0.7 0.7856 0.9154
CCC ext > 0.85 0.6546 0.6245
r2m > 0.5 0.3439 0.2544
Δr2m < 0.2 0.0805 0.3954
K’ 0.85 < k or k’ < 1.15 0.9775 0.9814
K 0.85 < k or k’ < 1.15 0.9984 0.9970
Table 3. Contribution of the descriptors and their predicted activity - Model 2.
Compound code Dataset split ATSC4m AATSC7i GATS5c SpMax5_Bhm hmin Exp. activity pIC50 Model predicted activity
A1 Excluded -150.613 0.16999 1.42508 3.125674028 0.025416 -1.19866
A2 Training -566.95 0.148565 1.322089 3.315003422 0.013312 -0.71767 -0.7177
A3 Training -363.531 0.300571 1.245976 3.139123587 0.037135 -0.65418 -0.6542
A4 Training -1158.25 0.10946 1.347909 3.448229765 0.012974 -0.76042 -0.7604
A5 Training -1849.53 0.063419 1.388731 3.451318505 0.01317 -0.84011 -0.8401
A6 Training -434.373 0.189907 1.338616 3.29591003 0.033856 -0.93852 -0.9385
A7 Training -144.952 0.077896 1.379928 3.142732176 0.013698 -0.65128 -0.6513
A8 Training -473.824 0.215142 1.086649 3.151668164 0.020063 -0.38382 -0.3838
A9 Training -170.971 0.426573 1.004151 3.417292915 0.039043 -0.86213 -0.8621
A10 Training -275.883 0.16498 1.091155 3.135368648 0.029323 -0.42325 -0.4233
A11 Prediction -351.663 0.109021 1.331711 3.200803294 0.018424 -0.63749 -0.6375
A12 Training 293.3892 0.222251 1.264795 3.341759912 -0.01408 -0.77597 -0.776
A13 Training -327.75 0.083413 1.33318 3.210689641 0.024596 -0.89209 -0.8921
A14 Prediction -673.156 0.110354 1.287059 3.389979292 0.002123 -0.7185 -0.7185
A15 Training -422.907 0.212385 1.365634 3.151249129 -0.00161 -0.70927 -0.7093
A16 Training -182.257 0.026798 1.323677 3.143652882 0.012183 -0.40654 -0.4065
A17 Excluded -238.246 0.147417 1.409326 3.165794266 0.00683 -0.42975
A18 Training -294.377 0.188816 1.128388 3.445302924 -0.03249 -0.17609 -0.1761
A19 Training -416.36 0.052274 1.241908 3.133696389 0.045083 -0.62737 -0.6274
A20 Excluded -212.48 0.185326 1.266641 3.352860035 0.023835 -1.09621
A21 Training -222.585 0.327564 1.194323 3.284275227 0.039924 -1.05308 -1.0531
A22 Prediction -102.236 0.082769 1.200738 3.453240814 0.033808 -0.59879 -0.5988
A23 Training -804.624 0.091836 1.40347 3.303221314 -0.00211 -0.54283 -0.5428
A24 Training -436.695 0.106619 1.328013 3.147730981 0.0188 -0.48855 -0.4886
A25 Training -331.053 0.229521 1.136152 3.303653226 0.004253 -0.35218 -0.3522
A26 Training -511.048 0.166717 1.106404 3.313512999 0.037071 -0.59218 -0.5922
A27 Training -497.33 0.179447 1.112946 3.302582849 0.013513 -0.31175 -0.3118
A28 Prediction -495.748 0.090511 1.20676 3.125788268 0.011527 -0.45939 -0.4594
A29 Training -119.239 0.184796 1.449173 3.301426063 0.009606 -1.16554 -1.1655
A30 Training -129.674 0.080919 1.422394 3.13276872 0.040722 -1.0607 -1.0607
A31 Training -132.681 0.211961 1.454825 3.450396768 0.009165 -1.22634 -1.2263
A32 Training -212.997 0.244178 1.456508 3.452942404 0.009421 -1.26007 -1.2601
A33 Prediction -208.635 0.179124 1.401204 3.179870948 0.016408 -0.8704 -0.8704
A34 Training -158.595 0.061056 1.208048 3.493150991 -0.17347 0.9914 0.9914
A35 Training -481.208 0.237308 1.17508 3.68518807 -0.23563 1.236572 1.2366

The statistical metrics supporting the model’s reliability: a low Friedman’s LOF of 0.0185 suggests minimal overfitting, and the multivariate correlation index Kxx of 0.3240, along with a global descriptor correlation Delta K of 0.0683, signifies a strong correlation between the descriptors and response variables. The model also exhibited a high F-statistic value of 172.9347, confirming the robustness of the regression analysis. These parameters collectively reflect the model’s high performance. Figures 2, Figure S2, and Figure S3 provide graphical representations of the experimental and predicted activity, residuals, and Z values, illustrating the model’s overall predictive accuracy.

Figure S2

Figure S3

Figure S4
Scatter plot of experimental and predicted activity of model 2.
Figure 2.
Scatter plot of experimental and predicted activity of model 2.

3.1.3. Internal validation of the model

Model 2’s internal validation indicates its reliability and predictive strength, as evidenced by cross-validated correlation coefficients obtained through both LOO and LMO methods, yielding Q2loo and Q2lmo values of 0.9551 and 0.9463, respectively. The Q2loo value reflects the model’s strong internal predictive power. The difference between the R2 and Q2loo is 0.0212, indicating that the model’s stability and not overfitting (Figures 3 & S4). Cross-validation error parameters also highlight the model’s precision, with a low Root Mean Square Error (RMSEcv) of 0.1177 and mean absolute error (MAEcv) of 0.0946, indicating accurate predictions that closely align with the observed values. The Predictive Error Sum of Squares (PRESS cv), a measure of total prediction error, with a value of 0.3740, further affirms the model’s predictive accuracy. Additionally, the CCCcv of 0.9775 suggests a strong agreement between predicted and observed values in the cross-validation process. A Y-scrambling test confirms the model generated is not due to a random chance, as evidenced by the parameters having low R2Yscr (0.1933) and Q2Yscr (-0.4139) values (Figure S5). These parameters are significantly lower than the original model’s R2 and Q2. This difference verifies that the model’s performance is robust and not obtained due to chance correlation.

Figure S5
Internal validation leave-one-out (LOO) plot of the model 2.
Figure 3.
Internal validation leave-one-out (LOO) plot of the model 2.

3.1.4. External validation of the model

The external validation of Model 2 reveals its predictive accuracy and precision on data outside the training set, as indicated by key metrics. The external validation regression coefficient R2ext is 0.5011, stating that the model explains approximately 50.11% of the variance within the external dataset. This shows the model’s moderate ability to generalize to new data. The external Q2ext metrics, including Q2ext-F1 (-0.0681), Q2ext-F2 (-0.4163), and Q2ext-F3 (0.9154), provide further insights into the model’s external predictivity. The high value of Q2ext-F3, >0.6, suggests better predictive strength. The model’s prediction errors, including the root mean square error (RMSE ext) of 0.1616, Mean absolute error (MAE ext) of 0.1285, and PRESS ext of 0.1305, show relatively low error values, supporting the model’s accuracy. Lower error values indicate that the model has higher predictive capability, as predictions were closely aligned with observed results. Moreover, the CCC ext of 0.6245 further reflects the model’s precision by quantifying the agreement between predicted and observed activities in the external validation set. The model’s robustness is also indicated by r2m metrics: the r2m average of 0.2544 reflects its mean predictive power across randomizations, while the Δr2m of 0.3954 suggests a small discrepancy between randomized and actual performance. The external validation parameters also included the slopes k and k′, which assess the agreement between observed and predicted activities. For Model 2, the values were within the acceptable range (0.85 < k or k′ < 1.15), with k = 0.9970, and k′ = 0.9814. These results indicate strong linear correlation and minimal systematic deviation between predicted and actual activities. These validation parameters confirm that Model 2 can have reasonable predictive power and good robustness on external data. The external predictive ability can be improved by certain modifications to the structure of the compounds.

3.1.5. Statistical representation of the model

The predicted QSAR model 2 exhibits strong fitting characters and validation parameters. QSARINS software provides four graphical representations that provide more insight into the model generated by the statistical results.

3.1.5.1. Scatter plot

The evaluation of Model 2 through graphical analysis demonstrates its predictive accuracy and reliability, as provided in Figure 2. A scatter plot comparing experimental and predicted activity values shows that most compounds fall close to the regression line, indicating a strong alignment between predicted and observed values, which supports the model’s predictive power.

3.1.5.2. Residuals vs. predicted endpoint

A plot of predicted values against residuals reveals a random distribution of residuals. Residuals are the differences between observed and predicted values. This randomness indicates that the model does not display systematic bias, thus supporting its reliability for generating accurate predictions in new datasets (Figure S2).

3.1.5.3. Z value vs. residuals

A Z-value is a measure that indicates the standard deviation of a data point or a model coefficient from the mean. The plot of Z values versus residuals examines the relationship between standardized Z values and residuals. This graph displays a trend upward, suggesting that certain descriptors introduce variability in predictions, as reflected in higher residuals. This implies the significance of each descriptor’s regression coefficient and the residuals, assessing the model’s accuracy for specific descriptors (Figure S3).

3.1.5.4. Applicability domain

The applicability domain plot, known as the Williams plot (Figure 4), displays the standardized residuals on the y-axis and leverage (HAT values) on the x-axis for each compound in the dataset. The dashed horizontal lines represent the threshold of ±2.5 standard deviations for the residuals, which helps identify potential outliers. The vertical line at leverage h* = 0.667 threshold marks the boundary for identifying influential data points. Most compounds fall within the acceptable limits for both standardized residuals and leverage, suggesting that the model performs robustly across the dataset without being influenced by compounds. However, compound A22 stands out with either high leverage or extreme residuals, indicating that it may influence the model’s predictivity and robustness. Compound A22 has 4-bromo and 2-fluoro substitutions that produce strong electron-withdrawing effects and high polarizability, leading to atypical topological and electronic descriptor values. Thus, it is affecting the model’s SAR, reducing predictive reliability.

Applicability domain - William’s plot of the model 2.
Figure 4.
Applicability domain - William’s plot of the model 2.

The Insurbia graph illustrates the leverage values (HAT values) on the x-axis and the predicted values generated by the model on the y-axis. It assesses the influence of each compound on the model’s predictions and identifies potential outliers. The leverage threshold is found at h* = 0.667. The compounds are clustered on the lower left side of the plot, within acceptable leverage, indicating that they do not impact the model’s predictions. However, compound A22 with high leverage was observed as an outlier. These points ensure that the compounds do not skew the model’s results (Figure S6).

Figure S6

3.1.6. Principle component analysis of the model

The principal component analysis (PCA) plot illustrates the distribution of data points across the first two principal components (PC1 and PC2). The PCA plot helps in visualizing the spread and grouping of the data, aiding in identifying potential clusters or anomalies within the dataset. PC1 and PC2 explain 34.54% and 29.21% of the variance in the dataset, respectively. The PCA score plot determines that each data point represents a compound plotted according to its score on these two principal components. The compounds cluster in specific regions, showing that most compounds are located around the center of the plot, with a few compounds, such as A5, A9, A34, and A35, dispersed further from the cluster. Compound A5, bearing a 4-iodo substituent, exhibits increased molecular weight and polarizability, impacting hydrophobic and electronic descriptors. A9 contains a 4-difluoromethoxy group, contributing strong electron-withdrawing effects and influencing charge-based properties. A35, substituted with 3-chloro and 4-(OCH2C6H4-3-SO2CH3), introduces significant steric and electronic complexity due to its bulky sulfonylated aryl ether moiety. These structural variations result in distinct descriptor profiles, which explain their dispersion in PCA space and suggest potential deviations in model prediction behavior (Figure 5).

PCA analysis - score plot of the model 2.
Figure 5.
PCA analysis - score plot of the model 2.

The PCA loading plot of the descriptors influencing the model contains the principal components PC1 and PC2 with 34.54% and 29.21% of the variance, respectively. The descriptor components involved are the GATS5c, SpMax5_Bhm, hmin, ATSC4m, and AATS7m. The orientation and length of the arrows indicate the contribution and influence of each variable on the principal components. The components SpMax5_Bhm and hmin align strongly with PC1 and PC2, respectively, in opposite orientations to each other. The components ATSC4m and AATS7m are closely correlated with each other and align with PC1. AATS7m and GATS5c are less correlated with each other and strongly influence the PC1 and PC2 components, respectively. This plot provides insights into the contribution of the descriptors to each other and helps to identify patterns in variable relationships (Figure 6).

PCA analysis - loading plot of model 2.
Figure 6.
PCA analysis - loading plot of model 2.

3.1.7. Mechanistic interpretation

In QSAR models, mechanistic interpretation explains each molecular descriptor’s correlation with a specific biological activity or chemical property. This type of interpretation helps to reveal the molecular mechanisms or structural features contributing to the observed biological effects. Model 2 consists of five descriptors: ATSC4m, AATSC7i, GATS5c, SpMax5_Bhm, and hmin. These descriptors provide insights into molecular mass distribution, electronic properties, and structural compactness influence the biological activity of the compounds.

3.1.7.1. Autocorrelation descriptors - ATSC4m, AATSC7i, GATS5c

Centered Broto-Moreau Autocorrelation, lag 4, weighted by mass (ATSC4m) and Average Centered Broto-Moreau Autocorrelation, lag 7, weighted by ionization potential (AATSC7i) descriptors belong to a class of autocorrelation descriptors that stand for the distribution of a particular property. The property shown by the numerical value 4 or 7 refers to the lag or the number of bonds over which the autocorrelation is calculated. ATSC4m demonstrates the distribution of atomic mass within the molecular structure that is four bonds apart. This descriptor influences the molecule’s shape, size, and ability to interact with biological targets with a minimal value of -0.0002. AATSC7i describes the autocorrelation averaged and weighted by ionization potential rather than mass. Ionization potential is an indicator of the molecule’s electron-donating or accepting tendencies, which affect chemical reactivity and binding with biological targets. AATSC7i influences the electron distribution affected by the ionization potential, with a substantial negative coefficient of -1.0450. Geary Autocorrelation, lag 5, weighted by charge (GATS5c) descriptor focuses on the spatial arrangement of atoms, with weighting by specific atomic properties. GATS5c measures the autocorrelation of atomic charges at a lag of 5 bonds. This provides information on the charge distribution within the molecule that influences electrostatic interactions with an impact on decreasing pIC50, as shown by its coefficient of -1.8843.

3.1.7.2. Burden matrix descriptor - SpMax5_Bhm

The highest Eigenvalue of the Burden Matrix, weighted by atomic mass, path length 5 (SpMax5_Bhm), is derived from the Burden Matrix, which captures information about a molecule’s connectivity and atomic properties. The descriptor is weighted by atomic mass and path length 5, indicating the maximum length of atomic paths considered in the calculation. This descriptor correlates the molecule’s 3D shape, size, and bonding properties potential for interactions with biological targets. It influences the activity with a coefficient of -1.1046.

3.1.7.3. Hydrogen - Electronic state descriptor

The minimum hydrogen electronic-state (hmin) descriptor captures information about the electronic state of the hydrogen atoms within a molecule. The minimum H E-State reflects the lowest electronic state among the hydrogen atoms in the molecule. It has a coefficient of -9.0433, suggesting a strong negative relationship with activity.

3.1.7.4. Correlation matrix of Model 2

The correlation matrix among the descriptors used in the model, including ATSC4m, AATSC7i, GATS5c, SpMax5_Bhm, and hmin, states the relationship between them (Table 4). Each value is the degree of correlation between pairs of descriptors, with values ranging from -1 (perfect negative correlation) to 1 (perfect positive correlation), and 0 indicating no correlation. ATSC4m and AATSC7i have a weak positive correlation of 0.275, indicating minimal shared variability. GATS5c is weakly negatively correlated with both ATSC4m (-0.070) and AATSC7i (-0.368), suggesting little to no linear relationship with these descriptors. SpMax5_Bhm shows a mild negative correlation with ATSC4m (-0.206) and a slight positive correlation with AATSC7i (0.210). It has a stronger negative correlation with hmin (-0.595), implying that as SpMax5_Bhm increases, hmin tends to decrease significantly. These low correlation values between most pairs suggest limited multicollinearity, meaning the descriptors provide distinct information to the model. This distinctiveness in descriptor contribution is beneficial for robust model performance, as each descriptor independently contributes to the predictive power of the model without redundancy.

Table 4. Correlation matrix of the model 2
Factors ATSC4m AATSC7i GATS5c SpMax5_Bhm hmin
ATSC4m 1
AATSC7i 0.27504 1
GATS5c -0.07034 -0.3683 1
SpMax5_Bhm -0.20583 0.209701 -0.15429 1
hmin -0.01133 0.028658 0.126401 -0.59526 1

3.1.8. Significance of QSAR model 2

The QSAR model exhibits robust performance, meeting both fitting criteria and validation parameters effectively. The high coefficient of determination (R2) indicates that the model explains about 97% of the variance in biological activity based on the molecular descriptors, demonstrating a strong correlation between descriptors and activity. The minimal difference between R2 and R2adj suggests that the model is not overfitted, as the addition of descriptors did not inflate its performance. Internal validation through LOO cross-validation also confirms the model’s predictive accuracy, as it maintains a high correlation even when individual observations are excluded one at a time. Low prediction error values further reinforce the model’s robustness and reliability. External validation also yielded a high R2ext, reflecting strong predictive power, and the external Q2 parameter exceeding 0.7 highlights the model’s significant predictive performance. In terms of applicability domain, the model was reliable, with only a single outlier identified, indicating a well-defined domain. The correlation matrix of the descriptors AATSC7i, SpMax5_Bhm, and hmin from Model 2 shows a positive linear correlation between them. Also, a strong correlation was observed between GATS5c and hmin. This correlation emphasizes that the descriptors effectively capture the variations in biological activity. Overall, these findings demonstrate that Model 2 is highly accurate, avoids overfitting, and possesses a robust predictive capability, making it suitable for predicting molecular activities in novel datasets.

3.1.9. SAR of the compounds by the predicted model

The SAR analysis for the quinazoline-based compounds revealed significant insights into how substitutions impact predicted IC50 values as per Model 2 (Figure 7). All compounds possess a basic benzyl-6,7-dimethoxy-N2-phenylquinazoline core. Based on Model 2, compound A35 demonstrated the highest predicted activity (pIC50 = 1.2366) due to chlorine and (3-(methylsulfonyl) benzyl) oxy groups at the 3rd and 4th positions on the aniline ring attached to quinazoline, while compound A32 showed the lowest activity (-1.2601) due to an iodine group in the 3rd position of the same ring. For A35, the electron-withdrawing group (EWG) substitution increased the negative contribution of descriptor ATSC4m, and the positive contribution of SpMax5_Bhm indicates the structural complexity of the molecule. Compound A32 has the highest contribution to the model by the GATS5c descriptor due to the electronic characteristics spaced five bonds apart.

Structural representation of the quinazoline moiety modification based on the results of QSAR model 2.
Figure 7.
Structural representation of the quinazoline moiety modification based on the results of QSAR model 2.

Compound A9 with difluoro methoxy at the 4th position of the aniline ring contributed inversely by the AATSC7i and GATS5c descriptors. These findings indicate that the contribution of electronic properties such as electronegativity and the molecule’s charge distribution is inversely proportional. From the results of Model 1, compounds A1, A17, and A20 were removed as outliers. The compound A1 had no substitutions, A17 with tri-substituted EDGs on the aniline ring, and A20 with di-substituted EWGs at the 3rd and 4th positions of the aniline ring in the quinazoline moiety. However, the compound A35 was modified with heavier groups on the methoxy hydrogen atom at the aniline ring and exhibited the highest predicted activity.

The influence of the substitutions on the descriptors of Model 2 is analyzed based on the results. These findings reveal four important points that could enhance the activity of the compounds.

3.1.9.1 Atomic mass

Higher atomic mass contributes negatively to ATSC4m but positively to SpMax5_Bhm, which enhances the significance of atomic weight on activity.

3.1.9.2. Electronic distribution

Both electron-donating groups (EDGs) and EWGs are vital for descriptors AATSC7i and GATS5c, suggesting that they help to stabilize the electronic environment of the quinazoline scaffold.

3.1.9.3. Structural complexity

Branching at the 2nd and 4th positions of quinazoline enhances the contribution of SpMax5_Bhm due to increased molecular complexity.

3.1.9.4. Hydrophobicity

Greater hydrophobicity negatively impacts the hmin descriptor, affecting improved activity in compounds with hydrophobic substituents.

3.1.10. Design of new compounds based on QSAR results

Based on these observations, 35 new compounds were designed. The newly designed compound structures have been provided in Table S1, and their descriptors with predicted IC50 value have been shown in Table 5. Sulfanilamide, a moiety containing both -SO2NH2 (EWG) and -NH (EDG), was introduced at the 2nd position of quinazoline. This substitution modulates the electron density by both donating and withdrawing electrons in different regions, influencing the quinazoline’s electronic properties. Bulky substitutions at the nitrogen on the 4th position increased atomic mass, with compounds AA9, AA15, AA21, AA5, and AA35 showing higher predicted activities. Compounds with two bulkier groups on the carbon attached to nitrogen showed higher activity than those with a single bulky group. Furthermore, replacing the methoxy (EDG) with hydroxyl (EWG) at the 7th and 8th positions of quinazoline increased predicted activity, like the compounds AA6, AA10, AA11, and AA18. Acetylation of sulfonamide groups in the compounds AA19, AA20, AA22, AA23, AA24, and AA34 also enhanced activity, particularly when hydroxyl groups were present at the 7th and 8th positions. However, comparing the compounds AA19, AA30, and AA31 states that benzylation decreased activity by approximately two-fold. These outcomes deliver the influence of strategic substitutions on activity and provide a framework for further refinement and optimization of quinazoline-based compounds. SAR of the quinazoline moiety based on the QSAR Model 2 has been shown in Figure 7.

Table S1

Table S3
Table 5. Designed compounds (AA1-AA35) with predicted IC50 value.
Compound code ATSC4 AATSC7i GATS5c SpMax5_Bhm hmin pIC50
AA1 -533.189 0.002509 1.036127 3.414456 0.065807 -0.64258
AA2 -453.439 0.130929 1.004045 3.420632 0.050501 -0.60068
AA3 -902.706 -0.07108 0.884068 3.421453 0.059191 -0.15315
AA4 -930.443 0.172437 0.896532 3.423452 0.062548 -0.45813
AA5 -766.352 0.085725 0.966534 3.425607 -0.01753 0.189515
AA6 -478.884 -0.10195 1.0266 3.685572 -0.10614 0.729143
AA7 -594.196 -0.00707 1.242763 3.685572 -0.13316 0.490134
AA8 -876.88 0.14887 1.173416 3.457652 -0.04455 -0.0352
AA9 -917.574 0.014504 1.171318 3.463448 -0.09143 0.534811
AA10 -835.869 -0.06454 0.96125 3.434964 -0.0644 0.78397
AA11 -1123.41 0.046041 0.84098 3.438811 -0.02414 0.58423
AA12 -1217.77 0.109234 1.011644 3.467792 -0.05117 0.427863
AA13 -806.959 0.127347 1.15717 3.466768 -0.05986 0.132273
AA14 -688.305 0.064945 0.946959 3.437319 -0.03283 0.357986
AA15 -1265.45 0.100006 0.943967 3.433921 -0.02852 0.407185
AA16 -1378.12 0.17246 1.138614 3.466196 -0.05555 0.195976
AA17 -1050.85 0.097151 1.134443 3.53273 -0.08666 0.424959
AA18 -931.646 0.0186 0.921157 3.521058 -0.05964 0.653597
AA19 -920.617 -0.1052 0.981909 3.436655 -0.06935 0.847398
AA20 -1218.76 -0.01532 0.867213 3.440462 -0.02909 0.660946
AA21 -1282.64 0.049189 1.026245 3.469432 -0.05612 0.519038
AA22 -973.259 -0.03001 1.174635 3.465141 -0.09638 0.629124
AA23 -667.177 -0.04028 1.240267 3.685572 -0.13811 0.588911
AA24 -583.29 -0.12617 1.04241 3.685572 -0.11109 0.790319
AA25 -1413.04 0.027021 0.964416 3.435919 -0.03347 0.517012
AA26 -1485.39 0.100196 1.144259 3.468007 -0.0605 0.325085
AA27 -1170.51 0.125093 1.269868 3.690064 -0.07521 -0.11281
AA28 -1062.62 0.050525 1.051115 3.687698 -0.04819 0.113956
AA29 -626.497 0.045927 1.080116 3.687673 -0.03719 -0.12252
AA30 -689.322 -0.06795 1.077253 3.687673 -0.08407 0.43835
AA31 -763.772 0.005465 1.30879 3.690064 -0.11109 0.181987
AA32 -440.895 -0.02046 1.377887 3.690064 -0.15283 0.391733
AA33 -338.17 -0.1048 1.140411 3.687673 -0.1258 0.665047
AA34 -1032.2 -0.02939 0.945538 3.521298 -0.06459 0.72243
AA35 -1119.5 0.045213 1.141008 3.533164 -0.09161 0.524894

3.2. Physico-chemical properties

The physicochemical parameters of the designed top 15 ligands have been presented in Table 6. The Log P values, reflecting the lipophilicity of the compounds, range from 3.29 to 6.65. Most of the compounds exhibit moderate to high lipophilicity, which is likely to support efficient membrane permeability. Among all, AA33 has the highest Log P (6.65), suggesting substantial lipophilic character, while AA25 exhibits the lowest (3.29), indicating comparatively reduced hydrophobicity. The TPSA values range from 128.81 Å2 for AA9 to 163.10 Å2 for AA20. Most compounds fall within a range that may allow for good permeability and oral bioavailability. Compounds with TPSA closer to 140 Å2 or below, like AA21 and AA23, may have enhanced permeability. The molecular weights (MW) of the ligands range between 463.52 (AA10) and 601.69 (AA33), aligning with the typical MW range for orally active drugs, though higher MWs in compounds like AA33 and AA23 may slightly reduce bioavailability. HBAs and HBDs are within favorable ranges for drug-like properties, with HBAs ranging from 9 to 11 and HBDs between 2 and 4, which could enhance interaction with target biomolecules. Additionally, the number of rotatable bonds, ranging from 5 (AA18) to 9 (AA33), suggests an adequate balance between flexibility and rigidity. Compounds with a moderate number of rotatable bonds, such as AA18 and AA19, may benefit from sufficient conformational adaptability without excessive entropy loss upon binding. The combination of these parameters suggests that these ligands may possess optimal drug-like properties with potential for bioavailability and effective interaction with biological targets.

Table 6. Physico-chemical parameters of the top 15 designed compounds.
Compound code Log P TPSA MW HBA HBD No of rotatable bonds
AA6 4.87 150.80 511.56 9 4 6
AA 9 4.77 128.81 491.57 9 2 8
AA 10 4.16 150.80 463.52 9 4 6
AA 11 3.71 160.03 479.52 10 4 6
AA 18 4.84 150.80 499.55 9 4 5
AA 19 3.89 153.87 505.56 10 4 7
AA 20 3.45 163.10 521.55 11 4 7
AA 21 4.06 141.12 549.61 11 2 9
AA 22 4.51 131.88 533.61 10 2 9
AA 23 5.22 131.88 581.65 10 2 9
AA 24 4.61 153.87 553.60 10 4 7
AA 25 3.29 153.87 497.56 10 4 6
AA 33 6.65 136.80 601.69 9 4 9
AA 34 4.57 153.87 541.59 10 4 6
AA 35 5.19 131.88 569.64 10 2 8

3.3. ADME studies

The ADME profiling of the top 15 designed compounds indicates a generally favorable pharmacokinetic and safety profile, suitable for anticancer drug development (Table 7). All compounds exhibit low GI absorption, which may limit oral bioavailability but is acceptable for non-oral or targeted delivery approaches, particularly in hepatic cancer, where liver-targeted drug accumulation is often desirable. None of the compounds are predicted to cross the BBB, minimizing the potential for central nervous system side effects, which is beneficial for therapies intended for peripheral cancers.

Table 7. ADME parameters of top 15 designed compounds.
Compound GI absorption BBB permeant P-GP substrate CYP1A2 inhibitor CYP2C19 inhibitor CYP2C9 inhibitor CYP2D6 inhibitor CYP3A4 inhibitor LOG KP (skin permeation) cm/s
AA6 Low No No No Yes No No No -6.03
AA 9 Low No No No Yes Yes No Yes -6.28
AA 10 Low No No Yes No No No No -6.57
AA 11 Low No No Yes No No No No -7.02
AA 18 Low No No No Yes No No No -6.24
AA 19 Low No No Yes Yes Yes No No -6.70
AA 20 Low No No Yes No Yes No Yes -7.15
AA 21 Low No No No Yes Yes No Yes -6.86
AA 22 Low No No No Yes Yes No Yes -6.41
AA 23 Low No No No Yes Yes No Yes -5.86
AA 24 Low No No No Yes Yes No No -6.15
AA 25 Low No No Yes No Yes No Yes -6.98
AA 33 Low No No No Yes No No No -5.23
AA 34 Low No No No Yes Yes No No -6.37
AA 35 Low No No No Yes Yes No Yes -6.08

All the compounds are non-substrates of P-glycoprotein (P-gp), a transporter commonly associated with drug efflux and multidrug resistance. This suggests that these molecules are less likely to be expelled from cancer cells, thereby enhancing intracellular retention and therapeutic efficacy. In terms of metabolism, most compounds selectively inhibit CYP2C19 and CYP2C9, with fewer compounds affecting CYP1A2 or CYP3A4, and none inhibiting CYP2D6. This selective inhibition profile reduces the risk of broad-spectrum cytochrome P450-mediated drug-drug interactions, making these candidates metabolically manageable.

Skin permeation values (Log Kp) across the series fall within a moderate range (-5.23 to -7.15 cm/s), indicating suitable lipophilicity and low dermal absorption risk. These ADME predictions support the potential of these quinazoline derivatives as safe and effective anticancer agents, with favorable metabolic compatibility, low CNS exposure, and minimal interaction with efflux transporters. These attributes make them strong candidates for further preclinical development.

3.4. Toxicity analysis

The toxicity results of the designed ligands (Table 8) reveal a promising safety margin, particularly when compared with the reference compound, Erlotinib, which exhibits multiple active toxicological properties. Erlotinib is predicted to be hepatotoxic (active, 0.78), immunotoxic (active, 0.91), mutagenic (active, 0.55), and cytotoxic (active, 0.75), raising concerns over its toxicity liabilities. Compounds such as AA6, AA9, AA10, AA11, and AA18 showed low probabilities for hepatotoxicity (around 0.5 or higher), indicating a reduced risk of liver toxicity. These compounds also showed favorable cytotoxicity profiles, with high inactivity scores (0.86-0.92), suggesting lower potential for general cellular toxicity. Likewise, compounds AA19, AA20, AA21, and AA23 demonstrated favorable mutagenic profiles, enhancing their potential as non-mutagenic agents. Additionally, compounds like AA19 and AA24 showed inactive immunotoxicity profiles, suggesting a minimal risk of adverse immune effects. Compounds like AA6, AA9, AA10, AA18, and AA24, which exhibit low hepatotoxicity, carcinogenicity, immunotoxicity, and cytotoxicity risks, stand out as strong candidates for drug development. These findings highlight their suitability for continued development, with minimal concerns for cellular or genetic toxicity.

Table 8. Toxicity score of the top 15 designed compounds.
Compound Hepatotoxicity
 Carcinogenicity
 Immunotoxicity
 Mutagenicity
 Cytotoxicity
Toxicity Probability Toxicity Probability Toxicity Probability Toxicity Probability Toxicity Probability
AA6 Ina- 0.53 Act- 0.61 Act- 0.67 Ina- 0.67 Ina- 0.87
AA 9 Ina- 0.57 Act- 0.61 Act- 0.67 Ina- 0.67 Ina- 0.87
AA 10 Ina- 0.56 Act- 0.60 Ina- 0.67 Ina- 0.64 Ina- 0.86
AA 11 Ina- 0.52 Act- 0.64 Act- 0.93 Ina- 0.58 Ina- 0.92
AA 18 Ina- 0.54 Act- 0.62 Act- 0.85 Ina- 0.65 Ina- 0.88
AA 19 Act- 0.52 Act- 0.52 Ina- 0.87 Ina- 0.73 Ina- 0.86
AA 20 Act- 0.55 Act- 0.51 Act- 0.79 Ina- 0.69 Ina- 0.91
AA 21 Act- 0.53 Act- 0.50 Act- 0.92 Ina- 0.68 Ina- 0.92
AA 22 Act- 0.52 Ina- 0.50 Act- 0.58 Ina- 0.69 Ina- 0.91
AA 23 Act- 0.53 Act- 0.50 Act- 0.58 Ina- 0.68 Ina- 0.92
AA 24 Act- 0.55 Act- 0.52 Ina- 0.57 Ina- 0.75 Ina- 0.86
AA 25 Act- 0.55 Ina- 0.50 Ina- 0.95 Ina- 0.75 Ina- 0.58
AA 33 Ina- 0.58 Act- 0.56 Ina- 0.61 Ina- 0.70 Ina- 0.77
AA 34 Act- 0.53 Act- 0.53 Act- 0.59 Ina- 0.74 Ina- 0.87
AA 35 Act- 0.53 Act- 0.50 Act- 0.93 Ina- 0.68 Ina- 0.92
Erlotinib Act- 0.78 Ina- 0.51 Act- 0.91 Act- 0.55 Act- 0.75

Act- = Active; Ina- = Inactive

3.5. Molecular docking results

The docking results of the top 15 designed compounds with target protein 2A4L revealed significant binding affinities, with docking scores ranging from -6.41 to -8.30 kcal/mol, as represented in Table 9. 2D and 3D interactions of the top 3 compounds with standard (Erlotinib) have been shown in Figure 8. Compounds AA19 and AA20 exhibit the highest binding affinities, with docking scores of -8.30 and -8.25 kcal/mol, respectively, than the reference compound Erlotinib and cocrystal, which have docking scores of -5.49 and -7.15 kcal/mol, respectively. Both AA19 and AA20 demonstrate robust interactions within the binding site, including key hydrogen bonds with residues such as ASP86, LEU83, LYS129, and GLU12, alongside Pi-alkyl interactions with VAL18, ALA144, and LEU134. These strong hydrogen bonds and Pi-alkyl interactions suggest a well-stabilized binding configuration, further reinforced by Pi-anion contacts with negatively charged residues like ASP145 and Pi-Pi stacking interactions with PHE80. Other compounds, such as AA18 and AA24, also display high binding affinities with scores of -7.38 and -8.02 kcal/mol, respectively. AA18 shows a balanced interaction profile with a combination of hydrogen bonds (ASP145, THR14) and Pi-alkyl interactions, while AA24 establishes both hydrogen bonds and Pi-cation interactions, indicating stable and diverse binding patterns. Compounds AA6 and AA10, with scores of -6.77 and -6.61 kcal/mol, engage in multiple hydrogen and Pi-sigma interactions, which provide moderate binding stability. Interestingly, while Erlotinib forms significant interactions through Pi-Pi stacking with PHE80 and multiple Pi-alkyl interactions, it lacks the extensive network of hydrogen bonds and Pi-anion interactions seen in AA19 and AA20. Also, similar to cocrystal, residues ASP86, LEU134, and VAL18 were involved in binding with compounds AA19 and AA20. These findings revealed the significance of specific polar and hydrophobic interactions, particularly hydrogen bonding, Pi-alkyl, and Pi-anion interactions, in enhancing binding affinity. Compounds AA19 and AA20, with their superior binding profiles, emerge as promising candidates for further development.

Table 9. Docking results of the top 10 designed compounds with 2A4L.
Compound code Docking score (Kcal mol-1) Number of interacting residues Amino acid and type of interaction
AA6 -6.77 8 LYS33, ASP86, 92,145 (H-bond); LYS88, 89 (Pi-Alkyl); GLU162 (Pi-Anion) GLY13 (Vanderwaals)
AA 9 -6.92 6 LYS20, 89, LEU298 (H-bond); ARG297 (Vanderwaals); ILE10 (Pi-sigma), HIS84 (Unfavourable acceptor-acceptor)
AA10 -6.61 9 ASP86, 92, 145 (H-bond); LYS33, 88, THR14, GLN131 (Pi-Sigma); LYS89, VAL164 (Pi-alkyl)
AA11 -6.75 8 GLN265, ASN272, THR182 (Hydrogen bond); HIS119, ILE186, 275, ALA277, 279 (Pi-Alkyl)
AA18 -7.38 10 ASP145, THR14, ILE10, LYS89 (H-bond); SER181 (Vanderwaals); VAL18,64, ALA31, 144, LEU134 (Pi-alkyl)
AA19 -8.30 10 ASP86, LEU83, GLN131, GLU12 (H-bond); VAL18, ALA144, LYS33, Leu134 (Pi-Alkyl); PHE80 (Pi-Pi T shaped) ASP145 (Pi-Anion)
AA20 -8.25 8 LYS129, LEU83 (H-bond); ASP86 (Pi-Anion); ALA144, VAL18, LEU134, LYS33 (Pi-Alkyl); PHE80 (Pi-Pi Tshaped)
AA21 -6.41 6 GLN131(H-bond), LEU134 (Pi-Sigma); ALA144, ILE10, 31 (Pi-alkyl); GLU81 (Vanderwaals)
AA22 -6.72 4 ASP86, 92, LYS88, (H-bond); LYS89 (Pi-alkyl)
AA23 -7.38 6 TYR180, ILE209, ASP235 (H-bond); VAL156 (Pi-Alkyl); ARG169 (Pi-cation); GLU208 (Unfavorable acceptor-acceptor)
AA24 -8.02 5 THR182, TYR180 (H-bond); TYR179 (Pi-Pi T shaped); ARG150 (Pi-cation), ARG122 (Pi-lone pair)
AA25 -7.59 7 ASP235, ILE209, GLU208, CYS177, VAL156, ILE173 (Pi-Alkyl); ARG169 (Pi-cation)
AA33 -6.65 6 ILE173, ALA149, ARG150 (H-bond); ARG126, LEU124 (Pi-alky); TYR180 (Unfavourable acceptor-acceptor)
AA34 -7.98 6

LYS33, THR14, ASP86 (H-bond); LYS88, 89

(Pi-alkyl); GLY13 (Vanderwaals)
AA35 -7.92 9 THR182, GLN265, ASN272 (H-bond); LYS278, ALA116, 277, 279, ILE186, HIS119 (Pi-alkyl)
Erlotinib -5.49 9 THR14, GLN131 (H-bond); PHE80 (Pi-Pi Stacked); VAL18,64, ALA31,144, PHE82 (Pi-alkyl); LEU134 (Pi-sigma)
Cocrystal (Roscovitine) -7.15 5 ASP86 (H-bond), LYS89 (pi-cation), LEU134 (pi-sigma), VAL18, ALA31 (alkyl)
2D and 3D interactions of the top three compounds with the standard (Erlotinib).
Figure 8.
2D and 3D interactions of the top three compounds with the standard (Erlotinib).

3.6. Molecular dynamics simulations

The molecular dynamics (MD) simulation of the CDK2-AA19 complex was carried out for 50 ns to investigate the structural stability, conformational flexibility, and interactions of the ligand AA19 with CDK2.

3.6.1. Protein-ligand RMSD

The protein-RMSD analysis of the CDK2 protein backbone was performed to evaluate its structural stability throughout the 50 ns molecular dynamics simulation. The RMSD initially shows 1.50 Å, reflecting a well-aligned and stable starting conformation. As the simulation progressed, the RMSD gradually increased to 1.75 Å, followed by a slight rise to 2.0 Å at 10 ns, indicating minor structural adjustments as the protein adapted to the simulation environment. Between 10 ns and 20 ns, the RMSD slightly decreased to 1.75 Å, showing a transient relaxation of the structure. At 25 ns, the RMSD increased again to 2.0 Å. A minor decrease to 1.75 Å was observed at 40 ns, though the system stabilized near 2.0 Å by the end of the simulation. These results revealed that the protein maintained an overall stable conformation, with RMSD values consistently ranging between 1.75 Å and 2.0 Å. The minor fluctuations observed are attributable to the inherent flexibility of loop regions, which do not significantly impact the structural integrity of the protein. This stability represents the reliability of the protein’s conformation in the simulated environment.

The ligand RMSD (AA19) analysis was conducted to assess its binding stability and conformational flexibility within the active site of CDK2 during the simulation. Initially, the ligand exhibited an RMSD of 1.2 Å, indicating a well-aligned and stable starting conformation. However, a sharp increase in RMSD was observed at 5 ns, peaking at 3.0 Å, reflecting initial rearrangements and adjustments of the ligand within the binding pocket. Following this, the RMSD decreased to a range of 1.8-2.4 Å, signifying improved stability, although intermittent spikes were noted, reaching 3.0 Å at 15 ns and 4.8 Å at 25 ns, possibly due to the ligand exploring alternate binding conformations. Beyond 25 ns, the Ligand RMSD fluctuated between 3.6-4.8 Å, suggesting continued conformational flexibility. By the end of the simulation (at 50 ns), the ligand RMSD stabilized within the lower range of 1.8-2.4 Å, indicating that it achieved a more stable and consistent binding state. These results highlight the dynamic nature of ligand binding, with AA19 undergoing structural adjustments while maintaining significant interactions within the active site. The Protein-Ligand RMSD has been shown in (Figure 9).

The RMSD pattern of protein ligand complex.
Figure 9.
The RMSD pattern of protein ligand complex.
3.6.2. Root mean square fluctuation

The RMSF analysis reveals the dynamic flexibility nature of CDK2. Most of the residues demonstrated RMSF values below 2.4 Å, showing the overall rigidity and structural stability of the protein. This consistency indicates that the core regions of CDK2 remain well-maintained and robust during the simulation, supporting its functional integrity. Slightly higher fluctuations were observed in the loop regions and terminal residues, as expected, since these areas typically exhibit greater flexibility, enabling them to contribute to essential conformational adjustments. These dynamic properties in specific regions enhance the adaptability of CDK2, ensuring its effective interaction with the ligand and its role in biological processes. The Protein RMSF pattern has been shown in Figure 10.

Protein RMSF pattern.
Figure 10.
Protein RMSF pattern.
3.6.3. Protein-ligand contacts

The protein-ligand interaction analysis of the CDK2-AA19 complex involved hydrogen bonding, water-mediated bridges, and hydrophobic interactions played crucial roles in ensuring a stable and dynamic binding environment for the ligand. The protein-ligand interaction analysis has been shown in Figure 11.

Protein-ligand interaction analysis.
Figure 11.
Protein-ligand interaction analysis.
3.6.4. Hydrogen bonding and water bridges

Hydrogen bonding emerged as a vital stabilizing force in the CDK2-AA19 complex. Asp145 formed strong and consistent hydrogen bonds with an approximate distance of 1.2 Å, while additional water-bridged interactions at 1.4 Å further reinforced the ligand’s binding stability. Similarly, THR165 maintained strong hydrogen bonds at a short distance of 0.4 Å, complemented by water bridges at 0.6 Å, showing its vital role in ligand anchoring. LYS129 also contributed significantly, establishing hydrogen bonds at 0.6 Å, supported by water-bridged interactions at 0.7 Å, which provided dynamic adaptability to the binding environment. Moderate hydrogen bonding was observed for THR14 and GLY12, with distances of 0.4 Å, further strengthened by water-mediated interactions. These residues, with their strong hydrogen-bonding capabilities, ensured sustained interaction with AA19 and facilitated the ligand’s stable positioning within the active site.

3.6.5. Hydrophobic interactions

Hydrophobic interactions also played a prominent role in the ligand’s stabilization. Residues such as LEU134, VAL18, ILE10, and TYR15 demonstrated consistent hydrophobic interactions throughout the simulation. These non-polar interactions created a favorable environment that shielded the ligand from solvent exposure, enhancing its affinity for the binding pocket. The hydrophobic contributions complemented the hydrogen bonding network, ensuring a balanced interaction profile and reducing the likelihood of ligand dissociation.

4. Conclusions

The QSAR analysis of quinazoline-based compounds using Model 2 has provided a robust understanding of the structural factors influencing biological activity, achieving high predictive accuracy with an R2 of 0.9763 and a cross-validated Q2 of 0.9551. Out of 480 descriptors, those such as ATSC4m, AATSC7i, GATS5c, and SpMax5_Bhm proved the most impactful structural relationship. Compounds with bulkier substituents, especially those featuring sulfanilamide, demonstrated enhanced predicted activity due to their balanced electronic distribution and increased molecular weight. Furthermore, substitutions at the 2nd and 4th positions on the quinazoline core effectively optimized electronic and hydrophobic properties, enhancing activity. The newly designed compound AA19 exhibited the highest predicted activity of 0.8473 and an optimal docking score of -8.30 kcal/mol against the CDK2 target. The molecular dynamics simulation showed that the RMSD, RMSF values for AA19 remained consistently low, indicating stable binding interactions and structural integrity throughout the simulation.

This study presents a systematic computational strategy for designing and evaluating quinazoline derivatives targeting hepatic cancer. However, we acknowledge certain limitations. The relatively small dataset of 35 compounds may constrain the model’s external predictive ability, despite strong internal validation. The external validation metrics suggest the need for further enhancement of chemical diversity in the dataset. Additionally, the study is based solely on computational predictions without experimental confirmation. While these results offer valuable insights for lead prioritization, in vitro and in vivo validation will be necessary to confirm the biological relevance and safety of the designed compounds. Future directions also include expanding the dataset, applying scaffold-hopping techniques to explore novel chemical space, and integrating synthetic accessibility analysis to guide practical compound development.

Acknowledgment

The authors extend their appreciation to Taif University, Saudi Arabia for supporting this work through project number (TU-DSPP-2024-180).

CRediT authorship contribution statement

Mohammed T. Alotaibi: In-silico study, interpretation, writing and revision.

Declaration of competing interest

The author declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

All data generated or analyzed during this study are included in this published article.

Declaration of generative AI and AI-assisted technologies in the writing process

The author confirms that there was no use of artificial intelligence (AI)-assisted technology for assisting in the writing or editing of the manuscript and no images were manipulated using AI.

Supplementary data

Supplementary material to this article can be found online at https://dx.doi.org/10.25259/AJC_759_2025.

References

  1. World Health Organization. [Last accessed on 2023 Feb 23]. Available online: www.who.int.
  2. , , , , , , , , , , , , . Global epidemiology of liver cancer 2022: An emphasis on geographic disparities. Chinese Medical Journal. 2024;137:2334-2342. https://doi.org/10.1097/CM9.0000000000003264
    [Google Scholar]
  3. National cancer institute. [Last accessed on 2023 Feb 23]. Available online: www.cancer.gov.
  4. , . Nanoscale strides: Exploring innovative therapies for breast cancer treatment. RSC Advances. 2024;14:14017-14040. https://doi.org/10.1039/d4ra02639j
    [Google Scholar]
  5. , , , , . Modeling the epidemic of nonalcoholic fatty liver disease demonstrates an exponential increase in burden of disease. Hepatology (Baltimore, Md.). 2018;67:123-133. https://doi.org/10.1002/hep.29466
    [Google Scholar]
  6. , . Analysis of drug combinations: Current methodological landscape. Pharmacology Research & Perspectives. 2015;3:e00149. https://doi.org/10.1002/prp2.149
    [Google Scholar]
  7. , , . Progress in adjuvant chemotherapy for breast cancer: An overview. BMC Medicine. 2015;13:195. https://doi.org/10.1186/s12916-015-0439-8
    [Google Scholar]
  8. , , , . Molecular therapies and precision medicine for hepatocellular carcinoma. Nature Reviews. Clinical Oncology. 2018;15:599-616. https://doi.org/10.1038/s41571-018-0073-4
    [Google Scholar]
  9. , , , , . Design, characterization, and DFT exploration of new quinazoline-N-substituted analogs: Anti-cancer activity and molecular docking insights. Journal of Molecular Structure. 2025;1322:140420. https://doi.org/10.1016/j.molstruc.2024.140420
    [Google Scholar]
  10. , , , , , , . Heterocyclic anticancer compounds: Recent advances and the paradigm shift towards the use of nanomedicine’s tool box. Molecules. 2015;20:16852-16891. https://doi.org/10.3390/molecules200916852
    [Google Scholar]
  11. , , , , , , . Synthesis, in silico and in vitro studies of novel quinazolinone derivatives as potential SARS-CoV-2 3CLpro inhibitors. Arabian Journal of Chemistry. 2024;17:105384. https://doi.org/10.1016/j.arabjc.2023.105384
    [Google Scholar]
  12. , , , , , , , , . Tyrosine kinase inhibitors: From rational design to clinical trials. Medicinal Research Reviews. 2001;21:499-512. https://doi.org/10.1002/med.1022
    [Google Scholar]
  13. . The 4-anilinoquinazoline class of inhibitors of the erbB family of receptor tyrosine kinases. Il Farmaco. 2001;56:51-56. https://doi.org/10.1016/s0014-827x(01)01026-6
    [Google Scholar]
  14. , , , , , , , , , . AG337, a novel lipophilic thymidylate synthase inhibitor: In vitro and in vivo preclinical studies. Cancer Chemotherapy and Pharmacology. 1996;37:509-517. https://doi.org/10.1007/s002800050422
    [Google Scholar]
  15. , . Therapeutic progression of quinazolines as targeted chemotherapeutic agents. European Journal of Medicinal Chemistry. 2021;211:113016. https://doi.org/10.1016/j.ejmech.2020.113016
    [Google Scholar]
  16. , , , , . FDA drug approval summary: Erlotinib (Tarceva®) tablets. The Oncologist. 2005;10:461-466. https://doi.org/10.1634/theoncologist.10-7-461
    [Google Scholar]
  17. , . Aurora kinase inhibitor ZM447439 blocks chromosome-induced spindle assembly, the completion of chromosome condensation, and the establishment of the spindle integrity checkpoint in Xenopus egg extracts. Molecular Biology of the Cell. 2005;16:1305-1318. https://doi.org/10.1091/mbc.e04-10-0891
    [Google Scholar]
  18. , , , . Cyclin dependent kinase (CDK) inhibitors as anticancer drugs. Bioorganic & Medicinal Chemistry Letters. 2015;25:3420-3435. https://doi.org/10.1016/j.bmcl.2015.05.100
    [Google Scholar]
  19. . Cyclin-dependent kinases. Genome Biology. 2014;15:122. https://doi.org/10.1186/gb4184
    [Google Scholar]
  20. , . Breast cancer treatment: A review. JAMA. 2019;321:288-300. https://doi.org/10.1001/jama.2018.19323
    [Google Scholar]
  21. , , , , , , , , , , , , . Discovery of N-(4-(3-isopropyl-2-methyl-2 H-indazol-5-yl)pyrimidin-2-yl)-4-(4-methylpiperazin-1-yl)quinazolin-7-amine as a novel, potent, and oral cyclin-dependent kinase Inhibitor against haematological malignancies. Journal of Medicinal Chemistry. 2021;64:12548-12571. https://doi.org/10.1021/acs.jmedchem.1c00271
    [Google Scholar]
  22. , . Cell cycle, CDKs and cancer: A changing paradigm. Nature reviews. Cancer. 2009;9:153-166. https://doi.org/10.1038/nrc2602
    [Google Scholar]
  23. , , , , , , , , , , , , , , . Quinazolines as cyclin dependent kinase inhibitors. Bioorganic & Medicinal Chemistry Letters. 2001;11:1157-1160. https://doi.org/10.1016/s0960-894x(01)00185-8
    [Google Scholar]
  24. , , , , , , , , , , , . Design, synthesis and evaluation of quinazoline-chalcone hybrids as inducers of cell-cycle arrest and apoptosis in breast cancer via DNA damage and CDK2/ATR inhibition. European Journal of Medicinal Chemistry Reports. 2025;13:100250. https://doi.org/10.1016/j.ejmcr.2025.100250
    [Google Scholar]
  25. , , , , , , , , . 2-Aminoquinazoline inhibitors of cyclin-dependent kinases. Bioorganic & Medicinal Chemistry Letters. 2005;15:3881-3885. https://doi.org/10.1016/j.bmcl.2005.05.131
    [Google Scholar]
  26. , , , , , . Design, synthesis, molecular docking, and evaluation of sulfonyl quinazoline analogues as promising liver cancer drugs. Bioorganic Chemistry. 2024;153:107777. https://doi.org/10.1016/j.bioorg.2024.107777
    [Google Scholar]
  27. . PaDEL-descriptor: An open source software to calculate molecular descriptors and fingerprints. Journal of Computational Chemistry. 2011;32:1466-1474. https://doi.org/10.1002/jcc.21707
    [Google Scholar]
  28. , , , , , , , , , , , . Synthesis of novel N4-substituted and C2-disubstituted 1,4-benzothiazine-1,1-dioxide derivatives: Integrative computational strategies for breast cancer therapy. Journal of Molecular Structure. 2025;1338:142310. https://doi.org/10.1016/j.molstruc.2025.142310
    [Google Scholar]
  29. , , , , , , , . In-silico studies of pyrazolopyranopyrimidine as a potential anticancer inhibitor: Synthesis, network pharmacology, ADMET prediction, molecular docking, and dynamics simulations. Journal of Molecular Structure. 2025;1343:142829. https://doi.org/10.1016/j.molstruc.2025.142829
    [Google Scholar]
  30. , , , , , , , . Exploring the cyclization of thiosemicarbazone to 1,3,4-thiadiazole: Synthesis, characterization and in-silico study. Journal of Molecular Structure. 2025;1322:140385. https://doi.org/10.1016/j.molstruc.2024.140385
    [Google Scholar]
  31. , , , , , , , , , . Thymol-1,2,3-triazole derivatives: Network pharmacology, molecular simulations and synthesis targeting breast cancer. Journal of Molecular Structure. 2025;1321:140060. https://doi.org/10.1016/j.molstruc.2024.140060
    [Google Scholar]
  32. , , , , , , , . Synthesis, X-ray diffraction, and computational studies of acyclovir and HBG analogs derived from Triazolyl-1,4-benzothiazine and their oxidized forms for breast cancer and SARS-CoV-2. Computational Biology and Chemistry. 2025;118:108498. https://doi.org/10.1016/j.compbiolchem.2025.108498
    [Google Scholar]

Fulltext Views
693

PDF downloads
843
View/Download PDF
Download Citations
BibTeX
RIS
Show Sections

Suggested read for related articles: