QSAR, molecular docking and ADMET studies of quinoline, isoquinoline and quinazoline derivatives against Plasmodium falciparum malaria

With the aim of researching new antimalarial drugs, a series of quinoline, isoquinoline, and quinazoline derivatives were studied against the Plasmodium falciparum CQ-sensitive and MQ-resistant strain 3D7 protozoan parasite. DFT with B3LYP functional and 6-311G basis set was used to calculate quantum chemical descriptors for QSAR models. The molecular mechanics (MM2) method was used to calculate constitutional, physicochemical, and topological descriptors. By randomly dividing the dataset into training and test sets, we were able to construct reliable models using linear regression (MLR), nonlinear regression (MNLR), and artificial neural networks (ANN). The determination coefficient values indicate the predictive quality of the established models. The robustness and predictive power of the generated models were also confirmed via internal validation, external validation, the Y-randomization test, and the applicability domain. Furthermore, molecular docking studies were conducted to identify the key interactions between the studied molecules and the PfPMT receptor’s active site. The findings of this contribution study indicate that the antimalarial activity of these compounds against Plasmodium falciparum appears to be largely determined by four descriptors, i.e., total connectivity (Tcon), percentage of carbon (C (%)), density (D), and bond length between the two nitrogen atoms (Bond N–N). On the basis of the reliable QSAR model and molecular docking results, several new antimalarial compounds have been designed. The selection of drug candidates was performed according to drug-likeness and ADMET parameters.


Introduction
Despite the high cost assigned to it, malaria remains one of the major diseases affecting the most vulnerable populations, with more than 500,000 deaths per year worldwide [1][2][3]. Among the various Plasmodium parasite species, Plasmodium falciparum is the most prevalent. The absence of a safe and effective vaccine and the emergence of antimalarial drug resistance are significant public health concerns. To ascertain the primary factors underlying malaria resistance, efforts must be directed toward developing novel inhibitor compounds capable of specifically targeting and inhibiting Plasmodium falciparum parasite growth and transmission [4][5][6][7]. Phosphatidylcholine (PC) is a key phospholipid component of parasite membranes and is required for parasite growth, proliferation, and survival. As a result, it is believed that phospholipid biosynthesis pathways represent promising targets for antimalarial drug development. In the case of malaria, the parasites require active PC generation via a Abdellah El Aissouq, Samir Chtita, and Fouad Khalil contributed equally to this work. different mechanism, the serine decarboxylation pathway (SDPM), in order to rapidly proliferate intraerythrocytic and gametocytes [8]. The unusual synthesis of phosphocholine from ethanolamine is catalyzed by phosphoethanolamine methyl transferase (PMT). Thus, inhibiting the catalytic domain of Plasmodium falciparum's phosphoethanolamine methyl transferase (PfPMT) can completely shut down the SDPM pathway. Apart from the fact that there are no human orthologues (mammalian absence), PfPMT may be used as a target template for the development of novel antimalarial drugs [9]. As a result, the crystal structure of PfPMT 3D7 (PDB ID: 3UJ9) can be used to conduct molecular docking experiments [10].
To optimize the biological activity of the molecules studied via molecular docking, the most active and least active molecules were chosen as the key to discovering new antimalarial drug candidates. As a result, numerous molecules with potential biological benefits have been proposed. The leverage values effect hi and the h* threshold were also calculated in order to select designed molecules within the applicability domain. Then, based on their pharmacokinetic properties, the molecules of the hit drug candidates were chosen. Finally, we confirmed the interactions of the newly designed compounds with the PfPMT receptor active site using the validated molecular docking method [22].
Among 69 descriptors calculated, fourteen molecular descriptors have chosen to create QSAR models, some of which were linear and some of which were non-linear, and to identify spatial regions associated with the activity of the compounds under study.
After optimizing the energy of each compound using the MM2 method (force field method with gradient for root mean square (RMS) of 0.01 kcal/mol) [23], topological, physicochemical, and geometrical descriptors were calculated using the Chem3D V16 [24], ChemSketch12 [25], and Marvin Sketch [26] software packages. Then, using the Gaussian 09 program [27], we calculated quantum chemical descriptors by employing the DFT approach with the B3LYP functional and the 6-311G basis (refer to Table 2, table S1a and S1b) in order to accurately determine the descriptors that correlate with the quality of molecular geometry optimization.
These descriptors served as input data for establishing a quantitative relationship between them and antimalarial activity using MLR, MNLR, and ANN methods as well as statistical analysis.

Principal component analysis and data splits
Principal component analysis (PCA) is a statistical technique that uses descriptive statistics. In this study, the PCA was used to extract as much information from the database as possible and to identify the various chemical descriptors that will help build the QSAR models [28]. Following this, the database was split into training and test sets comprising 80% and 20% of the total data, respectively [29]. This approach is carried out using the k-means classification technique (implemented in the XLSTAT software) [30]. Following this division, a randomly chosen compound from each cluster is included in the test set.

Models' development and validation
We employed multiple linear regression (MLR) [31], multiple nonlinear regression (MNLR), and artificial neural X and Y: N, CH; R1 and R2: -H, -OCH3; n =0, 1, 2; Chain's position (p) =3, 4 Fig. 1 Structures of quinoline, isoquinoline, and quinazoline derivatives network (ANN) approaches to construct QSAR models [32]. The MNLR and MLR models were created using the XLSTAT V. 2019 software [33], whereas the ANN model was created using the MatlabV.2015 software [34]. The principal parameters utilized in statistical analysis of QSAR models are the determination coefficient ( R 2 ) (Eq. 1), adjusted coefficient ( R 2 adj ) (Eq. 2), mean squared error (MSE) (Eq. 3), Fisher's statistical parameter (F-value), and level of significance (p-value) [35][36][37]. where Y iobs is the value of the observed response i, Y cal is the value of the predicted response i, Y cal is the average value of predicted responses, p is the number of explicative variables in the model, and n is the number of molecules in the training set. Internal validation was conducted to ensure the developed models' validity. In this step, the validation procedures named leave-one-out cross-validation (LOOCV) and leave-many-out cross-validation (LMOCV), based on the calculation of the R 2 cv coefficient value, were used to validate the robustness of developed models.
where Y 2 jobs (train) is the value of the observed response; Y jcal (train) is the value of the response predicted by Loo-cv; (1) Y cal (train) is the mean value of the predicted responses.
The predictive power of generated models is determined by calculating the R 2 test coefficient between the observed and projected pIC50 values for the test set [41].
The Y-randomization test was used to exclude the possibility of random association between selected descriptors and their associated activities in the original model [42].
To ensure that the model was not obtained by chance via the Y-randomization test, the average random correlation coefficient ( R 2 r ) of the randomly constructed models must be less than the correlation coefficient ( R 2 ) of the original non-random model [43].

Applicability domain (AD)
The applicability domain is defined as a space including molecules with accurately predicted activities; because the model is based on a limited number of compounds, it does not encompass the complete chemical space [44,45].
Among the several ways for defining AD models, the most widely utilized is determining the leverage values effect hi for each compound (Eq. 5) [46]. If the leverage effect hi of a compound exceeds the alert leverage h* (Eq. 6), the compound is regarded to be outside the application domain [47].
where i = 1, 2,…, n, xi is a vector descriptive of the compound to be discovered, and X is a matrix containing the model's k descriptor values for the n compounds in the training set.

Drug-likeness and pharmacokinetics ADMET prediction
To find possible drug candidates, the created compounds' drug similarity is evaluated using Lipinski rule-based filters, synthetic accessibility, and an in silico analysis of adsorption, distribution, metabolism, and excretion. The toxicity levels of the compounds under investigation were determined mostly during the drug research phase [48][49][50].
In this work, the drug-likeness and ADMET of the selected compounds are evaluated using the online Swis-sADME [51] and pkCSM [52,53] servers, respectively. Dipolar moment (DM), energy of highest occupied molecular orbital (E HOMO ), energy of the lowest unoccupied molecular orbital (E LUMO ), electrophilicity index (ω), bond length between two atoms of nitrogen inside chain (Bond NN ), angle between benzene carbon and nitrogen (Angle Bn-C-N), dihedral angle between benzene ring and carbon before nitrogen and nitrogen and carbon after nitrogen (DA)

Molecular docking
The PfPMT receptor is required for the catalytic production of PC. The amino acid His132 acts as a general base in the reaction mechanism of this receptor (Fig. S1), abstracting a proton from the hydroxyl group of Tyr-19 and activating the residue [10]. As a result, the crystal structure of PfPMT3D7 can be employed in molecular docking to deduce the critical structural requirements for anti-Plasmodium falciparum activity.
To conduct molecular docking analysis in this study, the AutoDock Vina tool was used. The Protein Data Bank has revealed the X-ray crystal structure of PfPMT3D7 (code PDB: 9UJ9). The Discovery Studio software [54] was used to visualize existing interactions in order to gain insight into the activity's critical structural requirements.
After removing the complex's original ligand and water molecules, the 3D grid maps were constructed using the AUTOGRID technique. The AutoDock software MGLTools 1.5.6 packages [55] was used to prepare the receptor and the original ligand (phosphocholine) for re-docking with the PfPMT receptor, as well as for active site identification using the same 3D grid [56]. To guarantee that the docking technique is acceptable and valid, the root mean square deviation (RMSD) range must not exceed 2 Å [57].

Dataset collection and selected descriptors
PCA method was applied using fourteen descriptors calculated for each quinoline, isoquinoline, and quinazoline derivatives (refer table S2). The descriptors with the lowest correlation coefficients are chosen and assigned to each of the 54 molecules analyzed in the form of a 13-column, 43-row matrix (refer table S3). The acquired database is then partitioned into training and test sets using the K-means technique as follows: the test set contains eleven molecules (1 h-1i-1i-1n-1p-1q-1p-3r-3v), while the training set contains forty-three molecules (refer table S4).

Models' development
Several models may be created utilizing specified chemical descriptors and experimental values of antimalarial activity. Any model that did not meet the Organization for Economic Cooperation and Development's principles, as well as Golbraikh and Tropsha's criteria, was discarded [58][59][60]. The equation for the resultant model produced using the MLR technique is provided below (Eq. 7): Because the model's p-value is less than 0.0001, we would take a risk of less than 0.01% assuming that H0 is incorrect and the model equation is statistically significant at a level larger than 95%.
According to this table, all descriptors chosen for the created model have VIF less than 5. This shows that there is no multicollinearity among the descriptors chosen and that the resulting model is stable.
The MLR model was tested for robustness by doing 100 random trials with randomized training set compound activity [43,62]. This technique generated 100 new models with new R r ,R 2 r , and Q 2 r values. Then, a comparison was done between the findings of the created model and the acceptable limit of Golbraikh and Tropsha's threshold values in order to corroborate  Table 4 explains how the optimal model's reliability and acceptability are determined [50]. The MLR model's coefficient of determination ( R 2 = 0.734) indicates strong activity-descriptor relationship efficiency (73%). Because the constructed model has excellent descriptive capacity to descriptors, the high adjusted coefficient of determination ( R 2 adj ) shows the genuine impact of employed descriptors on the analyzed pIC50. The cross-validated square correlation coefficient, for leave-one-out technique ( Q 2 LOOCV = 0.676) and leave-many(5)-out technique ( Q 2 L5OCV = 0.687), indicates that this model has a high degree of internal predictive capacity. The large value of R 2 test ( R 2 test = 0.664) indicates that the created model has a high degree of predictive capacity for the novel compounds. Additionally, the VIF, R 2 r , Q 2 r , and MSE values demonstrate that this ideal model may be used to forecast pIC50 values for novel quinoline, isoquinoline, and quinazoline compounds. The correlation diagram with calculated versus experimental pIC50 values of the MLR model of training and test sets is shown in Fig. 2.
Observing the distribution of observed and predicted pIC50 values, we can see that the two values are significantly correlated (refer to Fig. 2).
With the four molecular descriptors (TCon, C percent, D, and BondNN) as input parameters, MNLR and ANN models are developed to improve the relationship between predicted activities and these pertinent descriptors. The equation below (Eq. 8) illustrates the nonlinear model obtained via MNLR: where R 2 = 0.701, MSE = 0.121, and R 2 test = 0.690. When comparing experimental and predicted values of pIC50, the MNLR technique has a more consistent distribution of values (refer to Fig. 3).
To determine whether the descriptors chosen were effective at predicting pIC50 values, a feed-forward ANN model with the 4-4-1 architecture was used (refer to Fig. 4a). The sigmoid transport function in the hidden layer and the linear transfer function in the output layer form the basis of this technique. There are three layers of neurons in this ANN architecture: an input layer (4 neurons), a hidden layer (4 neurons), and an output layer (1 neuron). The number of neurons in an input layer is equal to or fewer than the number of descriptors and the output layer includes activity predictions. The number of hidden neurons in the hidden layer may be determined by calculating the parameter ρ = (number of weights)/(number of neurons). The value of this parameter should be 1 < ρ < 3 in order to guarantee that the ANN model is statistically acceptable [32]. As a result, the value 1.520 of the parameter ρ indicates that the number 4 in the hidden layer is proportional to the number of descriptors in the input layer in order to predict the pIC50 values expressed as 1 in the output layer. With a high value of the determination coefficient ( R 2 = 0.740), a low value of the mean square error (MSE = 0.087), and a high value of the test-validation coefficient ( R 2 test = 0.865), the ANN model was found to be highly effective in predicting the antimalarial activity of the investigated molecules. A similar even distribution of candidate pIC50 values was observed across the training and test sets, as illustrated in Fig. 4b. With this distribution, it is ensured that the predicted values of pC50 are extremely close to the values observed experimentally.
MLR, MNLR, and ANN models all exhibit statistical significance (refer table S6). As a result, these models can be used to forecast the biological activity of previously unknown drug candidates.

Applicability domain (AD)
To determine the appropriate application of the MLR model, researchers examined the relationship between residual value and leverage effect. The leverage effect threshold value h*, with h* = 3*(k + 1) /n); k = 4; n = 43 and the distribution of normalized residual values and leverage level values were calculated (refer to Fig. 5) [49]. From this diagram, compound which has hi > h*, with h* = 0.348, or with standardized residual greater than y = ± 3 is considered outside of the AD.
The MLR model predicts correct and valid pIC50 values for compounds that fall within the applicability domain, which is located to the left of the leverage threshold h* = 0.348, but not for molecules that fall outside of the applicability domain. As seen in this figure, one molecule 1 (1 h) belonging to the test set has a standard deviation outside the y range (y = − 3), while two molecules, one belonging to the training set's molecule 3 (1c) and the other to the test set's molecule 8 (3c), are outside of the domain of applicability. It is possible that the activity of these compounds was underestimated due to the lack of experimental data for these molecules. As a result, these Fig. 4 a ANN architecture (4-4-1). b Correlations between observed and the predicted activity using the ANN model molecules should be removed from the list of molecules and excluded from molecular modeling in the future study, which will focus exclusively on molecules within the applicability domain.

Molecular docking of most and less active molecules
In order to optimize the biological activity, molecular docking was used to compare the key interactions between the most active molecule 1c and the least active molecule 3q with the PfPMT protein receptor.
This visualization clearly shows that the catalytic dyad between Tyr19 and His132 is formed by phosphocholine binding sites.
The next phase will be to re-dock the phosphocholine with the PfPMT receptor for confirming the identity of its active site. This technique corroborates the validity of molecular docking as demonstrated in the following study. For this purpose, the grid maps were constructed using 68, 68, and 68 pointing in x, y, and z directions with grid point spacing of 0.375 A˚ and the center grid box is of 19.437 A˚, 11.576 A˚, and 17.437 A˚. Figure 8 depicts the conformational relationship between a docked ligand and its native crystallized form in the PfPMT receptor pocket, which is superimposed on the docked ligand. Figure 8 shows that the original and redocked ligands are almost perfectly superimposed in the PfPMT receptor pocket. A further point to note is the low value of the root mean squared deviation (0.256 A˚), which indicates that the grid maps and AutoDock software were effective in achieving excellent molecular docking results.
After identifying the active site involved in the inhibition of the PfPMT receptor, we perform validated docking with the PfPMT receptor for the most active molecule 1c and the least active molecule 3q using same grid maps (68, 68, and 68 pointing in x, y, and z directions with grid point spacing of 0.375 A˚ and the center grid box is of 19.437 A˚, 11.576 A˚, and 17.437 A˚). Figure 9 depicts the three-dimensional and two-dimensional visualizations of the interactions of the  As can be seen in Fig. 9a, the most active molecule 1c has conventional H-bonds between the nitrogen hydrogen chain and His132 (2.180 A°) and Lys180 (2.930 A°) and Asp 85 (2.750 A°). Carbon hydrogen bonds were formed with Tyr 19 as crucial amino acid and Arg127. Electrostatic interactions were formed with His132, Asp10, and Asp85 through Pi-ion bonding and formed Pi-sulfur interactions with Cys87. Both the benzene ring and heterocyclic ring packed performing hydrophobic interaction through pi-pi bonding with important residues like Tyr19 that is essential for biological function of PfPMT, while the least active molecule 3q (11.500 μM) is docked into this pocket by interactions with the following residues: Tyr19 through unfavorable acceptor-acceptor and pi-pi interactions, His132 and Asp85 through Pi-ion interactions, Cys87 by pi-sulfur interactions and through carbon hydrogen bond with His132, Tyr27, Ser64, and Ser239 (refer to Fig. 9b).
We have observed that the drug is most potent when the binding energy is lowest and there is the greatest number of interactions with the receptor. By comparing the binding energies of the ligand 1c (− 9.6 kcal/mol) and the ligand 3q (− 7.7 kcal/mol) with the PfPMT receptor, we can interpret the lowest experimental IC50 value for the active molecule 1c (0.032 μM) in comparison to the less active molecule 3q (11.500 μM).
Based on the results of the molecular docking predictions, it is clear that the structure of ligand 1c can be used to improve the inhibition of the enzymatic activity of the PfPMT protein, which has been demonstrated in this study. We can also modify the structure of molecule 1c and assess the impact of these modifications on the pIC50 values in order to develop new antimalarial molecules.
When looking for structural clues for new antimalarial activity, we can deduce that determining how long a carbon chain is between two amino functions on a side chain and its shape and atom types at the terminal amine will be key. Therefore, increasing antimalarial activity can be achieved by decreasing the shape or surface area of interacting molecules (for 3q PSA = 72; VWSA = 1082.31 and for 1c PSA = 43.43; VWSA = 806,08) as well as by decreasing the volume (increasing density D), increasing the percent of atoms carbon (C% = 77.3 for 1c and for 3q C% = 72), increasing total connectivity (Tcon) by increasing the number of hydrogen bond (NHB = 7 for 1c and NHB = 4 for 3q) and decreasing the bond length between two nitrogen atoms by decreasing the number of rotatable bond (NRB) or number of active torsion (12 for 1c and 15 for 3q).

Proposition of new compounds anti-Plasmodium candidates
As shown in the MLR model equation, steric characteristics of the substituents affect antimalarial activity; bond NN decreases activity, while total connectivity (Tcon), carbon percent ratio (C (%)), and density (D) increase activity.
The significance of each descriptor was determined by comparing its absolute value to the t-test (Student's t-test) or standardized coefficient; a larger t-test value indicates that the model's descriptor has a higher impact. The best model descriptors have T-values of 5.021, 5.757, 8.856, and − 1.396 for total connectivity, C (%), density (D), and bond length NN, respectively. The length of the bond between two nitrogen atoms (Bond NN) has a negative sign, implying that shortening it can increase activity (increase value of pIC50). The carbon atoms' percent ratio (C (%)), density (D), and total connectivity (Tcon) are all positive, indicating that increasing these values results in increased activity. As a consequence, the models may be used to predict the biological activity of additional molecules formed by modifying the structure of studied derivatives. By substituting R1, R2, and R7 as well as other modifications to active molecule 1c, twelve derivatives were designed using the zinc-database as a free library of commercially available compounds for virtual screening for substitutions (refer to Fig. 10 and table S7) [63].
We have seen that molecules that lie beyond the application area of the MLR model are not considered drug candidates. Drug-likeness was also assessed by utilizing the SwissADME online tool, which includes the Lipinski rule of five and synthetic accessibility (SA), which is a crucial feature to examine in this selection process since profiteering may occur in the pharmaceutical industry. Table 5 shows the  Each of the twelve proposed compounds has a synthetic accessibility value between 3.24 and 8.79, which is greater than 1 but less than 10. Thus, all designed compounds were chosen as drug candidates based on their ease of synthesis.
According to values of h i that are less than h*, seven molecules (L6, L7, L8, L9, L10, L11, and L12) are selected as drug candidates with the highest biological activities. Moreover, the designed ligands L6 and L11 generally exhibited better antimalarial activities on comparison with the molecule 1c as the most active molecule in the studied series.
Based solely on the best anticipated pIC50 values of the seven compounds, it is difficult to favor one molecule over another and pick it as the greatest inhibitor of PfPMT activity. The pharmacokinetic parameters ADMET were employed through the pkCSM online tool to ensure that the identified compounds were viable drugs. As a result, we will only select compounds with drug-like characteristics. The results of the ADMET properties prediction are presented in Table 6.
If the absorbance is less than 30%, the absorption is low. However, the seven compounds with a value more than 90% have a good absorbance in the human intestine [64], indicating that they are well-absorbed. Several studies have shown that the values for large volume of distribution (VDss) exceed 0.45 [52]. As a result, all of the proposed compounds have the most significant potential in terms of volume of distribution.
In terms of blood-brain barrier (BBB) and central nervous system (CNS) permeability standard values, when LogBB < − 1, compounds are poorly distributed to the brain and when LogBB > 0.3, compounds have the potential to cross the BBB. Furthermore, if LogPS > − 2 compounds are considered to penetrate the CNS, while LogPS < − 3 is difficult to move in the CNS [65,66]. Thus, all proposed compounds have the best significant potential to cross the barriers.
The enzymatic metabolism of a medication in the body indicates its chemical biotransformation. CYP enzymes are found in all bodily tissues and oxidize foreign germs to aid in their elimination. When inhibitors of this enzyme impair its metabolism, the drug may have the opposite effect [67]. CYP3A4 is the most important inhibitor in this study among the CYP families (CYP1A2, CYP2C9, CYP2C19, CYP2D6, and CYP3A4), which are responsible for the biotransformation of more than 90% of drugs undergoing phase metabolism. This implies that all newly synthesized compounds must function as both a substrate and an inhibitor of CYP3A4 [68]. As a result, only the compounds L9, L10, L11, and L12 were chosen as CYP3A4 substrates and inhibitors.
Clearance measures how quickly drugs are excreted from the body in comparison to their concentrations inside. Drug persistence is not a problem for any of the new compounds in Table 6.
During the earliest stages of drug development, the toxicity study of the predicted compounds plays a significant role. Almost every drug in this research is being tested for toxicity using the AMES test [69]. Table 6 shows that all of the molecular designs are non-toxic. We have found that almost all of the compounds in this series are poisonous, including molecule 1c, which was previously identified as the most effective antimalarial candidate.
We conclude that compound L11 fits all of the pharmacokinetic parameters tested in this contribution research based on the findings of the ADMET characteristics. Thus, compound L11 may be employed as an antimalarial medication in the future by blocking the PfPMT protein's enzymatic activity. Additionally, this molecule may be exploited to develop novel drugs with enhanced anti-Plasmodium falciparum activity.
To acquire insight into the critical interactions between ligands and PfPMT, validated molecular docking was performed on all hit compounds. The interactions of the most active molecule in the studied series 1c with PfPMT were then compared to those of the best designed ligand L11.

Molecular docking of selected molecules as drug candidates
To facilitate docking analysis, all ligands were docked to the active site of the protein using a validated 3D grid (grid used in docking validation stage). Table 7 summarizes the binding energies and interactions of the ligands L6, L7, L8, L9, L10, L11, L12, and 1c (as a reference molecule) with the PfPMT receptor.
From Table 7, the best ligands L6 and L11 have the lowest binding energy (− 10.400 kcal/mol and − 9.910 kcal/mol respectively) compared to other ligands in the order (1c, L7, L8, L9, L10, and L12). Thus, by comparing the values of the binding energies of the best ligands L6 and L11 with the receptor PfPMT, we can interpret the highest pIC50 value of the designed molecules L6 (7.990) and L11 (7.510) that reflect the high activity of these molecules compared to the experimental pIC50 value of the molecule 1c (7.490). In addition, the stronger biological activity of L6 and L11 can be also explained by the number and type of interactions with the active site of receptor (refer Fig. S2).
Number of interactions with important amino acids (His132 and Tyr19) involved in PfPMT inhibition shows that these interactions are important for binding energy and biological activities. The ligands (L6, 1C, L11, L7, L10, and L12) with pIC50 values above 6.500 have the most interactions in PfPMT inhibition compared to L8 and L9. Therefore, the formation of a hydrogen bond with the amino acid His132 is also important in the inhibition of drug candidates against P. falciparum.
The primary hit ligand L11 also exhibited strong interaction within the binding pocket by occupying critical amino acids Tyr19 and His132 that functions as a general base to extract a proton(H +) from the hydroxyl group (OH) of Tyr19 to activate the residue (refer to Fig. 11).
The hydrogen atom linked to nitrogen atom formed conventional hydrogen bonds with crucial amino acid Tyr19 (4.64A°), Asn10 (3.10A°), Asn109 (3.09), and Asp128 (2.82A°). The carbon that is linked to nitrogen atom plays a main role of formation of carbon-hydrogen bonds with  -ASN137-TYR160-TYR181   ASP10-ASP36-ASP128 TYR19  ASP85-ASP85 L7 L7-PfPMT − 9.600 HIS132-GLY63 LYS180-ASP128-ARG127 1C-PfPMT − 9.500 HIS132-LYS180-ASP85 TYR19-ARG127 TYR19 ASP10 Fig. 11 The 3D interactions (a) and 2D interactions (b) of the best designed ligand (L11) crucial amino acid His132 (3.37A°), Tyr181 (3.19A°), Lys180 (3.77A°), ILE36 (2.97A°), Asp128 (3.58A°), and Gly63 (4.89 A°). The benzene and heterocyclic rings play a crucial role of hydrophobic and electrostatic interactions. According to the docking results for selected ligand L11, we can conclude that the drug-like properties against Plasmodium falciparum may be improved by increasing the number of hydrogen bonds in proposed ligands in order to avoid the liberation of toxic ions due to the stronger interactions between these ligands and the PfPMT receptor. Thus, it is possible that hydrogen bonds suppressed in the chemical graph structure for molecular connectivity calculations could be considered a key parameter in this research. As a result, they should be included in any discussion of molecular connectivity indicators for new antimalarial drug candidates [70].
We notice that the Tcon descriptor establishes a direct relationship between the number of interacting hydrogen bonds (C-H and X-H) and the molecular connectivity index for each molecule. These results corroborate the QSAR and molecular docking analyses, as well as the experimental results.
Based on the results that are achieved by combining the study of 2D-QSAR with ADMET and molecular docking studies, it is clear that the structure of ligand L11 can be used to improve the inhibition of the enzymatic activity of PfPMT protein. In addition, we can modify the structure of this ligand in order to design new antimalarial drugs that can be added to the quinoline, isoquinoline, and quinazoline derivatives.

Conclusion
In this contribution study, molecular modeling was performed to improve the antimalarial activity against the Plasmodium falciparum CQ-sensitive and MQ-resistant strain 3D7 protozoan parasite of a set of fifty-four quinoline, isoquinoline, and quinazoline derivatives.
The statistical results obtained through internal and external validations corroborate the generated models' performance. According to the best MLR model, it is clear that steric parameters have the greatest effect on the investigated activities. As a result, the proposed models can be used to predict the antimalarial activity of novel quinoline, isoquinoline, and quinazoline derivatives.
To optimize the studied activity, the most and least active molecules were chosen based on the developed models' predictions. Then, docking study was conducted on these molecules with the PfPMT receptor in order to use them as a starting point for the development of new molecules with the highest activity.
Based on the obtained results, twelve molecules (L1-L12) with the highest antimalarial activity were designed. The results of applicability domain, drug-likeness, and ADMET properties showed that only the compound L11 had acceptable pharmacokinetics properties. Then, using molecular docking, the inhibitory activity of the best drug candidate L11 was confirmed. Due to its ability to interfere with the enzymatic activity of the PfPMT protein, this compound may be proposed as a novel antimalarial agent for use in the treatment of malaria. Moreover, it may also provide a wealth of opportunities for medicinal chemists to develop new Plasmodium falciparum drug candidates.