Molecular Docking, 3D-QSAR, and ADMET Study of 1,2,3-Triazole Derivatives for Multi-Target Drug Discovery

Diabetes mellitus (DM) is a common metabolic disorder that is marked by chronic hyperglycaemia through an intrinsic abnormality of insulin secretion, insulin action, or both [1]. Type II diabetes mellitus makes up about 95% of all cases. It is often linked to insulin resistance and the gradual decline of β-cells [2]. This leads to poor glucose control. Postprandial hyperglycaemia, due to inadequate insulin response after meals, is among the key clinical manifestations of T2DM [1]. The global data estimates, the incidence of diabetes is accelerating at a fast pace, with more than 463 million cases in 2019 and projections anticipating the figures to rise up to 700 million by 2045 [2]. Lifestyle change, rising urbanization, obesity, and unhealthy diet are major determinants of the trend. Poorly controlled diabetes has potential severe complications such as nephropathy, neuropathy, retinopathy, and cardiovascular disease. One of the useful means to control postprandial hyperglycaemia is inhibition of enzymes like α-amylase and α-glycosidase that break down carbohydrates and are liable for cleaving complex carbohydrate to glucose [2, 3]. Inhibition of these enzymes will slow down the absorption of glucose, thus reducing the risk of subsequent increase in blood sugar levels after eating [3]. Among the different classes of heterocyclic compounds studied for their therapeutic interest, derivatives of 1,2,3-triazole have been reported to show high biological activity. Heterocycles of nitrogen, they are also highly recognized by their pharmacological diversity, showing antibacterial, antifungal, antiviral, anticancer, and antidiabetic activities. Notably, many studies have pointed towards their ability to serve as α-glucosidase inhibitors upon structural modification or hybridation with other pharmacophores [4]. Thirty-two 1,2,3-triazole compounds were selected and screened in the present study by employing computer-aided drug design (CADD) tools. Atom-based 3D quantitative structure–activity relationship (QSAR) modeling, molecular docking simulation, and ADMET profiling are carried out in the experiment. The goal was to select candidate lead compounds with higher α-glucosidase and α-amylase enzyme inhibitory activity and thereby make them useful for their application in antidiabetic therapy.

MATERIALS AND METHODS

QSAR STUDIES

Quantitative Structure-Activity Relationship (QSAR) analysis was carried out to evaluate the antidiabetic potential of 1,2,3-Triazole derivatives using Schrödinger software (version 13.4). Only 3D QSAR modeling was performed in this study. The chemical structures of the compounds were collected from previously published research articles and were used without any structural modification. These compounds were optimized, and molecular descriptors were automatically generated using the software. The 3D QSAR model was constructed utilizing descriptors obtained from steric and electrostatic fields. This approach enabled the identification of essential structural features responsible for antidiabetic activity, reducing the need for exhaustive experimental testing. The model that was developed underwent validation to confirm its predictive accuracy. Computational Details;-Schrödinger Maestro v13.4, Excel for activity conversion.

1)  Data Set - 32 compounds a congeneric series of 1,2,3- Triazole derivatives with biological activity against α-amylase and α-glucosidase were selected for model generation with reported biological activity values (IC50)[15] Table no. [1]. The IC50 values were converted to pIC50 (- log(IC50)) for modeling. The data collected on the compounds demonstrated their significant effectiveness in exhibiting anti-diabetic properties. Among from which training set as 23 compounds and test set of 9 compounds to validate the model's predictive capability.

2) Molecular Alignment - Molecules were aligned using a common substructure-based alignment method in Maestro v13.4 (Schrödinger). Alignment is crucial for 3D-QSAR since spatial orientation significantly influences descriptor calculation. The molecular structures were drawn by Chem Draw Ultra 8.0 and converted into mol format for further studies.Template-based alignment was done using the most active compound as a template. Alignment performed in Maestro using flexible alignment.

3) Descriptor calculation - The development of the three-dimensional quantitative structure-activity relationship (3D QSAR) model was accomplished using the Phase module within Schrödinger Maestro 13.4. Ligands were methodically arranged according to their shared pharmacophoric features. The software independently generated three-dimensional field-based descriptors, which included steric, electrostatic, and hydrophobic properties. These descriptors were subsequently utilized to construct a partial least squares (PLS) regression model, which establishes a correlation between molecular features and biological activity. [8]

4) Model Building (CoMFA/PLS)-Partial Least Squares (PLS) regression was employed for model building and Cross-validation (leave-one-out method) was used to determine the optimal number of components. Partial Least Squares (PLS) regression was employed for model building due to its robustness in handling multicollinearity and high-dimensional descriptor data typical of CoMFA models. In PLS, the original descriptor variables are transformed into a new set of orthogonal variables called latent variables or components, which capture the maximum covariance between the predictors (descriptors) and the response variable (pIC50). This technique enhances predictive performance and model interpretability. To determine the optimal number of components for the PLS model, cross-validation using the leave-one-out (LOO) method was conducted. This strategy builds the model with the remaining compounds while eliminating one molecule from the training set. This model is then used to estimate the activity of the excluded molecule.   Every molecule in the training set goes through this procedure once more. The cross-validated Q² value, a key indicator of predictive accuracy, is computed from these predictions.

It is defined as: -Where, yi= the observed activity, y^i= the predicted activity from cross-validation, ? is the mean observed activity. A higher Q² (typically >0.5) suggests that the model has good internal predictive power.[16] The initial stage of QSAR modelling was creating a sufficient dataset (descriptors), which served       as the model's input. The compounds were separated into test and training sets in the following phase, data setup. According to the Kennard-Stone algorithm approach, 70% of the data was in the training set and 30% was in the test set.

5) Model Validation-The produced models were then subjected to internal and external validation, as well as fitting criteria validation. Internal validation was carried out using Y-scrambling (to confirm that the created model was not the consequence of random correlation), cross-validation leave-one-out (Q2LOO), cross- validation leave-many-out (Q2LMO), and root mean squared error (RMSE). External validation - The performance of the created model was then evaluated using various metrics, including RMSE external, Q2-F1, Q2-F2, Q2-F3, r2M, Δr2 m, and CCC. Statistical Parameters such as R² (coefficient of determination), Q² (cross-validated R²), RMSE (Root Mean Square Error), and Pearson correlation coefficient were calculated. Refer Table no. (2) and Fig no. (1)

6) Contour Map Visualization-Contour maps were generated to visualize the impact of different regions on biological activity; these maps guide the rational design of new derivatives by indicating which structural modifications could enhance biological activity. Refer fig.no. (2,3,4,5) Electron withdrawing Maps: Blue (Substitution allowed), Red (not favourable substitution). Negative Ionic Maps: Blue (favorable substitution for bulky groups) Hydrophobic Field Maps: Blue regions Favors hydrophobic groups. H-bond Donor Maps: Blue (favorable), Red (unfavorable).

Molecular Docking Studies

1) Molecular docking studies were conducted using Maestro (version 13.4, Schrödinger). A focused chemical library of 384 modified 1,2,3-triazole derivatives was created using compound 5 as the core scaffold from the 32 compounds screened for QSAR.

2) Initial 2D chemical structures were drawn using ChemDraw. Strategic substitutions were introduced at specific positions based on previous contour map analyses to improve binding potential.[10]

3) The prepared molecules were converted into 3D structures using Maestro’s 3D Builder Panel and processed using the LigPrep module. Ligands were standardized by correcting valency issues and adjusting ionization states to physiological pH (7.0 ± 0.0). Geometry optimization was performed using the OPLS_2005 force field to minimize energy until a root mean square deviation (RMSD) threshold of 1.8 Å was achieved. The lowest energy conformers were selected for docking.

4) Protein structures were retrieved from the Protein Data Bank: α-glucosidase (PDB ID: 1T69), a cancer-related protein (PDB ID: 2R3I). All proteins were pre-processed using the Protein Preparation Wizard, which included assigning bond orders, adding hydrogens and missing residues, optimizing protonation states, and minimizing structural clashes. Water molecules not involved in ligand binding were removed, while those contributing to interactions were retained. Receptor grid files were generated around the active sites of the native ligands using the GLIDE module. A cubic grid box of 14 Å per side was defined to ensure adequate binding site coverage for all ligands.[11]

5)  Docking was performed using the Extra Precision (XP) mode of Glide, utilizing enhanced scoring functions and stricter pose selection. Post-docking analysis focused on identifying key interactions including hydrogen bonding, hydrophobic contacts, π–π stacking, and van der Waals forces.[10]

In this study, all derivatives were designed based on a common 1,2,3-triazole-linked chromene-benzamide scaffold. The parent molecule comprised a chromene core bearing a hydroxyl group, linked through an ether chain to a 1,2,3-triazole ring, which was further connected to a substituted benzamide moiety. Systematic substitutions were introduced at specific positions on the terminal benzene ring to explore the effect on biological activity.[11]

ADME Studies

In silico ADMET predictions were carried out to forecast the activity of the synthesized triazole derivatives in the human body. The research used two reliable software programs: QikProp from the Schrödinger Suite (version 13.4) and the free Swiss ADME web server (https://www.swissadme.ch) [12]. All ligand geometries were generated using Schrödinger's LigPrep module. The 3D structures were then processed through QikProp, which provided various predicted properties relevant to drug development. These properties included molecular weight, number of hydrogen bond donors (HBD) and acceptors (HBA), partition coefficient (Log Po/w), polar surface area (PSA), solubility (Log S), intestinal permeability (QPPCaco) [22], ability to block hERG ion channels (QPlogHERG), and prediction of human oral absorption [23]. Compliance with Lipinski's Rule of Five was also checked to evaluate oral drug potential.ADME[Refer to table no 4 and 5]

Toxicity Studies

Toxicity studies are crucial steps in determining how an agent would potentially impact living organisms. These studies assist researchers in understanding whether a substance can be harmful, the type of harm it could cause, and at what concentration it becomes harmful. It aims to establish safe exposure levels for both animals and humans, particularly significant in drug development and chemical safety evaluations. Software Tools used to Assess Toxicity, to accelerate and streamline this process, researchers apply specialized software that can forecast how a compound will act within the body and if it could be toxic. Two of the tools frequently used are SwissADME and pkCSM.

SwissADME is an internet-based tool that assists scientists with assessing potential drug candidates. It forecasts the absorption, distribution, metabolism, and excretion of a compound in the body in unison referred to as ADME properties. It further evaluates whether a molecule appears to be "drug-like" or not based on its chemistry.SwissADME includes one of its features to create SMILES (a string notation of molecules), which are utilized further for evaluation, e.g., toxicity checking. This instrument streamlines the initial steps of drug discovery by identifying compounds that may be harmful or ineffective [15]. http://www.swissadme.ch/ [12].

pkCSM is another potent instrument that employs a method known as graph-based signatures to treat small molecules. Imagine taking an electronic fingerprint of a compound based on the arrangement and spacing of its atoms. This "fingerprint" is then applied to predict all sorts of pharmacokinetic properties, such as possible toxicity. pkCSM can predict how a compound may act in the body whether it may be toxic to organs, induce mutations, or have other hazards. It's especially useful because it enables researchers to eliminate toxic compounds early on, conserving time and resources [17] . https://biosig.lab.uq.edu.au/pkcsm/ [13]. [Refer to table no 6 and 7]

RESULTS AND DISCUSSION

In this research 3D QSAR studies were performed.

ATOM BASED 3D QSAR Studies

The 3D-QSAR models generated by PHASE are predicated on either pharmacophore or atomic frameworks. Given that all the ligands employed belong to a congeneric series, we opted for 3D-QSAR models based on atomic structures. [7] a dataset of 32 compounds was used using Schrodinger Software 13.4 ver. These were generated using the best 5-point HHPR (Hydrogen bond donar, Hydrophobic, Positive ,Aromatic ring)hypothesis. Default pharmacophore features in PHASE are hydrogen bond acceptor (A), hydrogen bond donor (D), hydrophobic (H), negative (N), positive (P) and aromatic ring (R), Other types (miscellaneous) (O)The hypothesis was generated using molecule training set and a grid spacing of 1.0 Q. 3D-QSAR model with three PLS factors was generated and further, validated with predicting activities of molecules in the test set.

1.Regression Analysis-Correlation plots of predicted vs actual activities for both training and test sets. Refer fig no. (6,7)

2.Conclusion-The 3D QSAR model showed good correlation between predicted and experimental values with high R² and Q². The 1-factor model - Low R² (0.35) and low Q² (0.29) indicate this model doesn't explain or predict activity well. The 3-factor model showed the best statistical parameters (R² = 0.83, Q² = 0.70), but its R² scramble value (0.56) indicated potential overfitting. Hence, the 2-factor model R² = 0.69, Q² = 0.39                    R2 scramble = 0.3968 considered more reliable than another factor.

Y- Randomization-The strength of the created models is surveyed using the Y-Randomization test. The dependent variable (-logIC50) is randomly mixed to construct a new QSAR model after each iteration. The model's low Q2 and R2 values show that the 3D-QSAR model that was created from the original data is strong and not based solely on conjecture. [9] The outcomes of 3 random shuffles for the Y-randomization test are displayed in Table (1) Q2 and R2 for CoMSIA that ranged from -0.2971 to 0.7085 and 0.356 to 0.835, respectively. This suggests that the created models are reliable and not based on random correlations. Contour maps provide a structural basis for future modifications to enhance activity. The model can guide rational drug design of 1,2,3-triazole derivatives against diabetes.

Newly designed compounds

According to the structure–activity relationship (SAR) analysis conducted in this study, the fifth compound from the aligned dataset exhibited the highest predicted biological activity. This lead compound served as the basis for structural modifications, which were subsequently docked against three key therapeutic targets: α-amylase and α-glucosidase (PDB ID: 1T69) for antidiabetic activity, cyclin-dependent kinase 2 (CDK2, PDB ID: 2R2I) for anticancer activity.

Fig no. (1) QSAR Analysis Output Showing Predicted Activities and Statistical Validation

Fig no. (2) Electron Withdrawing Map

Fig no. (3) Negative ion Field Maps

Fig no. (4) Hydrophobic Field Maps

Fig no. (5) Hydrogen Bond Donar Field Map

Fig no. (6) Correlation plot of predicted vs actual activities for both training set

Fig no. (7) Correlation plot of predicted vs actual activities for both test set

Table no (1) - QSAR dataset 32 compounds [25] a congeneric series of 1,2,3- Triazole derivatives with           biological activity against α-amylase and α-glucosidase were selected for model generation with reported biological activity values (IC50). The IC50 values were converted to pIC50 (- log(IC50)) for modelling.

Compounds	PHASE QSAR SET	Compound	IC50	PIC50	Predicted activity
1	Training		15.900	4.799	4.703
2	Training		11.820	4.927	4.882
3	Training		29.840	4.525	4.916
4	Test		2.060	5.686	5.063
5	Test		8.310	5.080	5.201
6	Test		2.780	5.556	5.241
7	Training		3.070	5.513	5.282
8	Training		6.130	5.213	4.994
9	Training		7.060	5.151	5.015
10	Training		5.230	5.281	5.175
11	Training		3.170	5.499	5.544
12	Training		17.610	4.754	4.983
13	Training		15.620	4.806	5.018
14	Training		5.870	5.231	5.171
15	Test		5.880	5.231	5.205
16	Test		98.630	4.006	4.044
17	Training		26.110	4.583	4.363
18	Training		32.330	4.490	4.390
19	Test		33.160	4.479	4.542
20	Training		42.600	4.371	4.666
21	Test		12.780	4.893	4.822
22	Training		12.130	4.916	4.929
23	Training		28.700	4.542	4.424
24	Training		37.400	4.427	4.405
25	Training		61.100	4.214	4.428
26	Test		27.400	4.562	4.390
27	Test		41.000	4.387	4.415
28	Test		29.400	4.532	4.478
29	Training		45.900	4.338	4.391
30	Training		48.400	4.315	4.383
31	Training		33.600	4.474	4.388
32	Training		28.200	4.550	4.432

Table no. (2) Statistical parameters and its Threshold value

Docking Outcomes and Binding Analysis

Among the 384 synthesized triazole derivatives, compounds 1A and 2A emerged as lead candidates against α-glucosidase (PDB ID: 1T69), with XP Glide docking scores of −7.206 kcal/mol. These scores indicate stronger binding potential than the standard antidiabetic agent, Metformin with XP Glide docking score -2.63. The binding analysis revealed stabilization of these ligands through hydrogen bonds with LYS33, ALA32, ASP101 van der Waals forces, and potential hydrophobic bonds. Calculated binding energies, approximately −58.3 kcal/mol, further supported their stable accommodation within the enzyme’s active site.[10] In the case of the cancer-related protein (PDB ID: 2R3I), compounds 1B and 2B demonstrated the highest affinities, with docking scores of −8.206 and −8.045 kcal/mol, respectively which exhibits better and stronger interaction than the standard Anticancer agent Letrozole with XP Glide Docking Score -3.30. These compounds exhibited predominant hydrogen bondings with amino acids like GLU62, HIE84, ASP86 and salt bridges with LYS89, suggesting strong complementarity with the target site. The enhanced binding profiles of these derivatives suggest their potential for dual inhibition, acting on both α-glucosidase and cancer-associated proteins.[11]

In summary, the docking studies identified compounds 1A and 2A as promising α-glucosidase inhibitors with superior performance to the reference compound. Additionally, compounds 1B and 2B demonstrated potential multi-target activity warranting further biological validation. Refer Table no.( 3)

Table No (3) Docking Score and Glide energies of top docked compounds

Structure	Naming	Docking Score	Glide Energy
	1A	-7.206	-58.326
	2A	-6.957	-50.943
	3A	-6.842	-55.654
	4A	-6.826	-57.782
	5A	-6.785	-63.015
	1B	-8.206	-73.3554
	2B	-8.045	-70.554
	3B	-7.798	-73.123
	4B	-7.488	-75.256
	5B	-7.439	-71.310

The optimal docking interaction produced by compounds 1A and 1B with protein 1T69, 2R3I respectively. Refer figure (6) and (7)

Fig 6. Interaction of 1A with IT69(2D and 3D)

Fig 7. Interaction of 1B with 2R3I(2D and 3D)

ADME

Table No (4) ADME Properties of Anti-Diabetic Compounds (PDB ID: 1T69)

Metformin included as standard reference. Abbreviations: HBD – Hydrogen bond donors; HBA – Hydrogen bond acceptors; PSA – Polar surface area; hERG – Human Ether-à-go-go-Related Gene inhibition; Log Po/w – Octanol/water partition coefficient; Log S – Aqueous solubility. SI units are used throughout.

Table No (5) ADME Properties of Anti-Cancer Compounds (PDB ID: 2R3I

2R3i included as standard reference. Same abbreviations as Table 4

The ADMET profiles of the triazole derivatives were assessed using QikProp and SwissADME to determine their potential for oral drug development [24]. Among the anti-diabetic candidates (PDB ID: 1T69), compounds 1A and 2A showed acceptable oral absorption levels between 10 and 100%. They also complied with Lipinski's rule. Compound 2A achieved an ideal balance of molecular weight (532.5 Da), PSA (189.5 Ų), and predicted aqueous solubility. Although some candidates went beyond the PSA or H-bond acceptor limits, all had good QPlogPo/w values ranging from 3.3 to 4.8, indicating good permeability [25]. In the anticancer group (PDB ID: 2R3I), compounds 1B and 2B stood out for their relatively low hERG inhibition, with values from –7.15 to –7.25, and their moderate predicted oral absorption. While most of these compounds had molecular weights over 600 Da, they still met reasonable drug-like criteria, with only two breaches of Lipinski's rule, which is common for kinase inhibitors. Their PSA values were acceptable, suggesting they could pass through cells.

Toxicity Studies

Table No. (6) TOXICITY STUDIES FOR ANTIDIABETIC COMPOUNDS [PDB-ID= 1T69]

If one considers the safety of putative Antidiabetic Compounds (PDB ID: 1T69) in terms of human health as opposed to environmental safety, compounds 2-A and 4-A are the safest. They do not induce AMES toxicity, indicating they will not produce genetic mutations. They do not block hERG I channels, which is significant in preventing heart-related side effects. Their highest tolerated doses in human beings and toxicity levels in rats (short- and long-term) are within safe limits. Compound 5-A, however, does have highest tolerated dose in human beings and does well in acute toxicity tests but it's AMES-positive, which is an issue concerning mutagenic effects. So, if we’re prioritizing non-mutagenic safety and a high safety margin, compound 4-A [CTB(BRCONH?).mol] comes out on top as the safest overall choice [17]. Metformin included as standard reference.

Table No. (7) TOXICITY STUDIES OF ANTICANCER COMPOUNDS [PDB-ID= 2R3I]