INTRODUCTION:

Diabetes has tremendous impact on the lives and welfare of people, families, and civilizations all around the world. Diabetes mellitus (DM) is a cluster of metabolic illnesses marked by sustained hyperglycemia, or a rise in glucose level in the blood, caused by a failure of insulin secretion, insulin action, or both.

Insulin, an anabolic hormone, is responsible for glucose, lipid, and protein metabolic abnormalities^1,2,3. The International Diabetes Federation (IDF) projects that the prevalence of diabetes would reach 9.3 percentage (463 million people) in the year 2019 and then increase to 10.2 percentage (578 million) and 10.9 percentage (700 million) by 2030 and 2045, respectively⁴.

Diabetes mellitus can lead to major complications such as cardiovascular disease, retinopathy, nephropathy, and neuropathy if left untreated for a long time⁵. These complications are sometimes exacerbated by co-morbidities such as obesity, cataracts, erectile dysfunction, and nonalcoholic fatty liver disease⁶. Current treatments include dietary adjustments, oral anti-diabetes medications, insulin injections, and incretin mimetics⁷. Due to poor glycemic control or unbearable side effects, the quest for novel, safe, and easy-to-administer options continues.

The NADPH-dependent reductase 11β-hydroxysteroid dehydrogenase type 1 (11-HSD1) enzyme has emerged as a major target for diabetes treatment. Dr. Monder of Rockefeller University cloned and purified 11-HSD1 in the 1960s. Carbazole sodium, Glycyrrhiza glabra root, and gossypol were discovered as the first non-selective 11-HSD1 enzyme inhibitors in the 1990s⁸. The interconversion of physiologically inert cortisone to active cortisol is controlled by the enzyme 11-HSD1. Excess glucocorticoid in liver, adipose, and skeletal muscle tissues may have an impact in the development of metabolic syndrome⁹. The 11-HSD1 genetically transformed mice were resistant to hyperglycemia caused by obesity or stress, according to studies. 11-HSD2, a type II isoform, catalyses the deactivation of cortisol with the help of the NAD cofactor. As a result, medicines that target 11-HSD1 have the potential to effectively manage blood glucose levels, improve insulin sensitivity, correct metabolic abnormalities, and modulate endocrine processes, making them viable diabetic treatments¹⁰.

Corwin Hansch and Toshio Fujita created QSAR analysis, a ligand-based drug design (LBDD) tool, more than 50 years ago (1964)¹¹. Using regression and classification approaches, it tries to establish a statistically important association between chemical compounds and biological activity (pEC50, Ki, pIC50 and so on). When compared to experimental testing, it has the advantages of higher rapidity and cheaper budgets for bioactivity evaluation^12,13.

Among the traditional methodologies, MLR and PLS are two widely used QSAR approaches. Regression study, on the other hand, implies a linear connection between biological activity and one or more than one descriptors^14,15. Biological processes, instead are nonlinear by definition, therefore the influence of some factors to a particular biological activity can also be nonlinear. Neural networks are the key to overcoming these challenges due to its nonlinear mapping¹⁶.

The research of QSAR (Quantitative Structure Activity Relationship) has become an expert practice that is applied as a significant and effective tool in drug development over time¹⁷. We used partial least squares, multiple linear regression, and artificial neural network procedures to create QSAR models on the same set of chemicals in this investigation.

MATERIAL AND METHODS:

Data set preparation:

The present QSAR research used a variety of 3-Oxazinan-2-one derivatives from the literature, as shown in (Table 1)¹⁸. The doses corresponding to 50% growth inhibition were determined experimentally with selected 11-hydroxysteroid dehydrogenase type 1 enzyme IC50 values. The reported inhibitory constant values were transformed to a negative log IC50 value¹⁹. Chem Draw Ultra 8.0 was fed into the TSAR 3.3 window via. mol files for all computational investigations. The seriesstructures were exposed to CORINA create 3D, which converts 2D molecules into 3D structures.The "COSMIC module of TSAR" was used to improve molecular shapes, and the charge 2 option was used to calculate charges²⁰. Because the IC50 values of some compounds were unknown, they were eliminated from the QSAR analysis.

Defining substitutions:

The definition of structure and replacements is an essential stage in the QSAR approach. Each chemical structures substituent was divided into two categories (namely R1 and R2). Every individual molecule had a specific number of substituents connected to the nucleus by single bond.²¹

Division of data set into two sets 1) Training set and 2) Test set:

The data is separated into two parts that is training set (Table 2) and test data set (Table 3) with structurally varied, active and inactive chemicals in each. The QSAR model is developed using training set, while test set is utilized to assess its predictability and accuracy. The training set QSAR model is validated using the test set. The model prediction is improved by using proper data separation techniques²². The overall number of structural compounds in training set was 31 while the overall number of structural compounds in the test set was 15. Five molecules (10, 11, 23, 42, and 47) were discovered that does not fit either of the training group or test group of compounds throughout the model construction and validation processes. Because these five compounds behaved like outliers, they were eventually removed from the training set.

Molecular descriptor:

Molecular mass, surface area, volume, dipole moments, lipole moments, molecular mass, moments of inertia, ellipsoidal volume, verloop parameters, Wiener index, molecular connectivity indices, molecular shape indices, electrotopological state indices, logP, number of defined atoms (carbon, nitrogen, etc.), rings (aromatic and aliphatic), and groups (methyl, hydroxyl, etc.)are among the descriptors that can be calculated. Total energy, electronic energy, atomic charge, mean polarizability, heat of creation, total dipole, nuclear repulsion energy, accessible surface area, polarizability, and dipole components were calculated using Vamp, a semi empirical molecular orbital module in TSAR 3.3.²³

More than 150 descriptors were calculated for the molecules of the series for QSAR studies as well as substituents depending upon the series using TSAR software.

Data reduction:

Only a few key descriptors are maintained for model construction out of a vast number of descriptors this can be accomplished using a data reduction strategy. The inter-correlation between two consecutive descriptors was verified using this method. If the correlation coefficient between the two descriptors was larger than 0.5, the descriptor having the lowest correlation with biological activity was removed. The one with a value of zero for all the molecules was likewise removed from the total calculated descriptors. Only descriptors that were strongly associated to biological activity but had no association between them were maintained after this approach was conducted for each pair of two successive parameters²⁴.

Multivariate statistical analysis:

The CADD employed in this drug development techniques are quite broad, and include equation-based models such as a) Multiple linear regression b) Partial least squares and non-equation-based c) Feed neural-network models.

Multiple linear regression analysis:

The ordinary least square regression approach is used in Multiple Linear Regression (MLR) study. MLR is a method for modelling a linear relationship between abiological activity which is denoted by dependent variable Yand Molecular descriptors represented by an independent variable X. MLR uses the least squares curve fitting approach to estimate the values of regression coefficients (r2). The model generates a linear (straight line) relationship the finest approaches of all the individual data points. The number of compounds in regression (n), degree of freedom, the regression coefficient (r2), Cross verified coefficient correlation (r2cv), and Fischer's test for statistical significance (F) were all employed to evaluate the models in the 2D QSAR^24,25.

Partial least square regression analysis:

The Partial Least Square regression model discovers new variables or latent variables by combining the original variables in linear combinations. When the data set comprises significantly inter-correlated descriptors (Multicollinearity) and when the number of descriptors surpasses the number of observations, the Partial Least Square is very relevant. The best number of Partial Least Square components (latent variables) for the study was discovered using a leave one out cross validation method. The Partial Least Square model was built using the same training data as the Multiple Linear Regression model^14,26

Feed-forward neural network (FFNN) analysis:

The researchers were inspired to create the FFNN model by the brain analogy. Neurons are the basic building blocks of the brain and have numerous connections. The FFNNs model is based on the idea that information is retained in the brain due to the strength of connections between neurons rather than the internal state of individual neurons. To improve the result, the parameters of the neural network models are tweaked. The power of FFNNs model is increased by combining techniques such as principal component analysis with other methodologies. The term "feed-forward" refers to the technique of using previous units' output as input to new units.The external inputs are represented by the units in the first layer, the outputs are represented by the units in the last layer, and the hidden units are represented by the units in the layers between the input layer (first) and the output layer (last)^16,27. In FFNN analysis biological activity is dependent on structural features of the molecules.The molecular descriptors (independent variable) are mapped to biological activity using the Monte Carlo technique (dependent variable). The FFNN model was built using the descriptors used in linear regression²¹

RESULT:

The MLR linear regression models were developed using 11β-hydroxysteroid dehydrogenase type 1 inhibitory activity (dependent variable) and the two descriptors (independent variables) namely total lipole (whole molecule), verloop B1 (subst.1) left after data reduction. Equations and statistical values used for the development of MLR model is below as equation (1)

Y = 2.0142045*X1 + 0.1873935*X2 – 5.2408509 (1)

Where, X1 = VERLOOP B1 (substitution 1), X2 =TOTAL LIPOLE (whole molecule)

s value: 0.280224, f value: 141.59, correlation coefficient r²: 0.95395, r^²: 0.91002 and r^²cv: 0.884137

The data above shows that the series has the highest correlation coefficient (r² 0.91), which explains 91% of the variance in biological activity, showing the model's fitness. There was also a small difference (0.03) amid r^² and r^² cv, indicating the soundness of the developed QSAR method. The cross validated by leaving out one row also showed a value of r^²cv of 0.88, which ideally should be higher than 0.6. The model's statistical relevance is further demonstrated by the larger value of f = 141.59 and small value of s value= 0.28 also testifies the statistical relevance of the model.The di parametric model is significant for describing the 11β - HSD1 inhibitory activity of the chemicals under investigation, based on the above data analysis. Partial Least Square regression analysis was done on the same data set to confirm robustness and prediction capacity of the developed MLR model. It is well knowledge that the results of MLR and PLS should be equivalent given a well-defined model. Equation is given by the PLS model that has been developed is below as equation (2)

Y= 2.0142045*X1 + 0.1873935*X2 – 5.2408509(2)

Where, X1 = VERLOOP B1 (substitution 1), X2 =TOTAL LIPOLE (whole molecule)

Statistical significance (1.0138), r^2cv (0.892552), fraction of variance r^2 (0.91002)

The statistical output of developed PLS model in terms of r^²cv (0.88867), statistical significance (1.042) and r^² value (0.89976) for test set compounds clearly shows that both the two descriptors identified during MLR model development are highly significant and are contributing heavily towards 11-β hydroxysteroid dehydrogenase type 1 enzyme inhibitory activity of the compounds under study.

Table 1: Structure of compounds with the given biological activity IC50 (nM) data of 1, 3-Oxazinan-2-one derivatives

Compound	Structure	R1	R2	IC50 (nM)
1		Me	3-Me	42
2		Me	3-Me	23
3		Me	3-Me	152
4		H	3-Me	999
5		H	3-Me	353
6		Me	H	160
7		Me	2-Cl	40
8		Me	3-Cl	103
9		Me	3-Br	23
10		Me	4-Cl	229
11		Me	C₆H₅- 4-CF₃	805
12		Me	H	631
13		Me	H	4.7
14		Me	2-F	12
15		Me	3-F	4.2
16		Me	4-F	3.3
17		Me	4-F	3.3
18		Me	4-F	29
19		Me	C₆H₅-2,4-di-F	1.6
20		CH₂=CHCH₂	C₆H₅-2,4-di-F	3.2
21		HOCH₂CH₂	C₆H₅-2,4-di-F	1.2
22		HOCH₂CH₂	C₆H₅-2,4-di-F	0.8
23		HOCH₂CH₂	C₆H₅-2,4-di-F	140
24		HOCH₂CH₂CH₂	C₆H₅-2,4-di-F	4.4
25		CH₂CH₂OH	H	0.7
26		CH₂CH₂CH₂OH	H	0.6
27		CH₂CH₂OH	H	0.5
28		CH₂CH₂CH₂OH	H	0.6
29		CH₂CH₂CH₂OH	H	0.6
30		CH₂CH₂OH	4-F	0.8
31		CH₂CH₂CH₂OH	4-F	1.1
32		CH₂CH₂CH₂OH	2-F	0.9
33		CH₂CH₂CH₂OH	3-F	0.9
34		CH₂CH₂CH₂OH	4-F	0.6
35		CH₂CH₂OH	4-F	0.5
36		CH₂CH₂OH	H	0.8
37		CH₂CONH₂	H	2.8
38		CH₂CH₂ NH₂	H	2.7
39		CH₂CH₂NHAc	H	7.1
40		CH₂CH₂NHCONH₂	H	0.8
41		CH₂CH₂NHCONHMe	H	3.1
42		CH₂CH₂NHCONHEt	H	22
43		CH₂CH₂NHSO₂Me	H	1.0
44		CH₂CH₂NHSO₂NH₂	H	2.0
45		CH₂CH₂CH₂OH	H	1.1
46		CH₂CH₂CONH₂	H	1.0
47		CH₂CH₂CONHMe	H	22
48		CH₂CH₂CH₂NH₂	H	5.7
49		CH₂CH₂CH₂NHCONH₂	H	1.4
50		CH2CH2CH2NHCONHMe	H	4.1
51		CH₂CH₂CH₂NHSO₂Me	H	1.8

Table 2: Actual activity and predicted activity data for compounds in training set

Compound Name	Actual activity (-log IC50 (nM)	Predicted activity (nM)
Compound Name	Actual activity (-log IC50 (nM)	MLR	PLS	FFNN
1	-1.62	-1.54	-2.21	-1.58
2	-1.36	-1.54	-2.03	-1.58
3	-2.18	-1.16	-2.28	-2.00
4	-2.99	-2.81	-3.01	-3.01
5	-2.54	-2.73	-2.97	-2.59
7	-1.60	-1.78	-2.45	-3.05
9	-1.36	-1.45	-1.78	-1.20
13	-0.67	-0.40	-0.44	-0.57
14	-1.07	-0.45	-0.48	-0.63
15	-0.62	-0.41	-0.45	-0.59
16	-0.51	-0.39	-0.42	-0.55
17	-0.51	-0.39	-0.40	-0.55
19	-0.20	-0.42	-0.45	-0.60
20	-0.50	-0.61	-0.69	-0.73
22	0.09	0.48	0.91	0.22
25	0.15	0.04	0.28	0.21
26	0.22	-0.06	0.10	0.17
27	0.30	0.05	0.20	0.21
28	0.22	-0.04	0.07	0.18
29	0.22	-0.04	0.14	0.18
30	0.09	-0.03	0.13	0.19
31	-0.04	-0.13	-0.03	0.06
32	0.04	-0.01	0.17	0.20
33	0.04	-0.06	0.06	0.16
34	0.22	-0.11	0.01	0.09
35	0.30	-0.01	0.17	0.19
36	0.09	-0.03	0.09	0.19
40	0.09	0.31	0.70	0.22
43	0	0.57	1.13	0.22
45	-0.04	-0.14	-0.04	0.05
46	0	0.06	0.31	0.21

Table 3: Actual activity and predicted activity data for compounds in test set

Compound Name	Actual activity (-log IC50 nM)	Predicted activity (nM)
Compound Name	Actual activity (-log IC50 nM)	MLR	PLS	FFNN
6	-2.20	-2.06	-2.38	-2.10
8	-2.01	-2.05	-2.42	-2.10
12	-2.80	-2.80	-3.04	-2.75
18	-1.46	-1.13	-1.17	-1.50
21	-0.07	-0.48	-0.37	-0.44
24	-0.64	-0.30	-0.21	-0.44
38	-0.43	-0.85	-0.85	-0.47
39	-0.85	-0.64	-0.61	-0.44
41	-0.49	-0.54	-0.46	-0.44
44	-0.30	-0.33	-0.19	-0.44
48	-0.75	-0.75	-0.73	-0.45
49	-0.14	-0.31	-0.22	-0.44
50	-0.61	-0.14	0.06	-0.44
51	-0.25	-0.32	-0.18	-0.44

FFNN evaluation were carried out with same descriptors that were used for MLR and PL Sanalyses in an attempt to further evaluate the importance of descriptors. For the present study the neural network run with 2 hidden nodes and training and test set generated the best value of correlation coefficient (r2). The r2 that is correlation coefficient of training and test set was found to be 0.907 and 0.935 respectively.

(Tables 2) and (Table 3) illustrateactual activity and predicted activity for training set and test sets of chemicals, respectively, based on Multiple Linear Regression (MLR), Partial Least Square (PLS) and Feed Forward Neural Network (FNNN) analysis.

The following (Figure 1), (Figure 2), (Figure 3), (Figure 4), (Figure 5), and (Figure 6) depicts the graphs plottedbetween actual activity and predicted activity of training set and testset obtained by MLR and PLS and FFNN respectively.

The following (Figure 7) and (Figure 8) show the dependency plots amongst biological activity and descriptors of training and test set used to build the model (A) Total lipole (whole molecule) and (B) Verloop B1 (Substitution 1).

Figure 1: Plot between actual activity on X-axies and predicted activity on Y-axies of training set MLR

Figure 2: Plot between actual activity on X-axies and predicted activity on Y-axies of test set MLR

Figure 3: Plot between actual activity on X-axies and predicted activity on Y-axies of training set PLS

Figure 4: Plot between actual activity on X-axies and predicted activity on Y-axiesof test set PLS

Figure 5: Plot between actual activity on X-axies and predicted activity on Y-axiesof training set FFNN

Figure 6: Plot between actual activity on X-axies and predicted activity on Y-axies of test set FFNN

(A) Total lipole (Whole molecule)

(B) Verloop B1 (Substitution 1)

Figure 7: Dependency plots between biological activity and descriptors of training set used to build the model (A) Total lipole (whole molecule), (B) Verloop B1 (Substitution 1)

(C) Total lipole (whole molecule)

(D) Verloop B1 (substitution 1)

Figure 8: Dependency plots between biological activity and descriptors of test set used to build the model (C) Total lipole (whole molecule) (D) Verloop B1 (Substitution 1)

DISCUSSION:

The results of MLR, PLS and FFNN analysis clearly shows the negative correlation of X1 (verloop B1) and positive correlation of X2 (total lipole).

Study of descriptors:

Verloop B1 (Substitution 1)

Verloop uses a computer program to create a multidimensional steric parameter based on standard van der wall radii, bond lengths and angles, and potential conformations for the substituent. These are also called as Sterimol programs. Verloop has seen extensive achievement in applying these parameters in QSAR studies. It contains sub parameters: width parameters and length parameter. B1, B2, B3, B4 are the width parameters, and L is the length parameter. The B1 parameter defines the smallest width of the substituent from the primary bond axis and B4 is the maximum width^28,29. The obtained QSAR model showed negative dependencies on the verloop B1 (substitution 1). The negative value shows that decrease in the width of the substitution will increase the biological activity.

Total lipole:

Second descriptor that come in the model is total lipole. Total lipole bear a resemblance to lipophilicity which is a ration of the capability of molecules to transfer between oily partition and aqueous partition. It is a term that is typically used to describe how easy a chemical can traverse physiological membranes. It is helpful in determining the pharmacokinetic parameters like absorption, distribution process with in the body systems which are estimated by the aqueous loving or aqueous hating properties of molecules. As lipophilicity is tough to determine directly, usually the water/octanol log P that is termed as partition coefficient is used to estimate it. It is premeditated from the summed atomic log P values, as dipole is premeditated from the summed partial charges of a molecule³⁰.

The obtained QSAR model showed positive dependencies on the total lipole (whole molecule) which shows that increase in the lipophilicity increases the biological activity.So increase in the lipophilicity decrease the 11β- HSD type 1 enzyme in the biological system.

CONCLUSION:

The 11β-hydroxysteroid dehydrogenase type 1 enzyme has appeared as a promising molecular aim for the advancement of anti-diabetic drugs. On the based 11- HSD1 inhibitors, QSAR models of 1, 3-Oxazinan-2-one derivatives were created using multiple linear regression (MLR), partial least squares (PLS), and feed forward neural network techniques (FFNN). All these models demonstrated the relevance of Verloop B1 (Substitution 1) and Total lipole molecular descriptor at most in predicting the anti-diabetic effectiveness of these inhibitors. The obtained QSAR model showed negative dependencies on the verloop B1 (substitution 1) and showed positive dependencies on the total lipole (whole molecule). As a result, the developed QSAR models will be helpful in the design and development of further novel 1, 3-Oxazinan-2-one derivativesas anti-diabetic 11-hydroxysteroid dehydrogenase type 1 inhibitors.

CONFLICT OF INTEREST:

The authors have no conflicts of interest regarding this examination.

ACKNOWLEDGMENTS:

Authors pay honest gratitude to Late Prof. Aditya Shastri, Vice Chancellor, Banasthali University, Tonk, Rajasthan, India for providing required computational conveniences for the accomplishment of the learning in a suitable way.

REFERENCE:

1. Kharroubi AT, Darwish HM. Diabetes mellitus : The epidemic of the century. 2015; 6(6): 850–67. doi: 10.4239/wjd.v6.i6.850

2. Karishma S, Lakshmi K, Tony DE, Babu AN, Nadendla RR. et al. Pharmacological Evaluation of Leaf Extract of Terminalia bellerica with Moringa oleifera for its Synergistic Action on Anti-diabetic Activity and Anti-inflammatory Activity in Rats. Res J Pharm Technol. 2019; 12(3): 1181-4. DOI: 10.5958/0974-360X.2019.00195.1

3. Sam SG, Kumar A, Babu VD, Swamy NV. et al. Evaluation of Anti-diabetic and Anti-inflammatory activities of ethanolic extract of whole plant of Inula racemosa. Research Journal of Pharmacology and Pharmacodynamics. 2015; 7(3): 129-34. http://dx.doi.org/10.5958/2321-5836.2015.00024.5

4. Saeedi P, Petersohn, Salpea P, Malanda B, Karuranga S, Unwin N, et al. Global and regional diabetes prevalence estimates for 2019 and projections for 2030 and 2045 : Results from the International Diabetes Federation Diabetes Atlas, 9 th edition. Diabetes Res Clin Pract. 2019; 157: 107843. https://doi.org/10.1016/j.diabres.2019.107843

5. Antony J, Debroy S, Manisha C, Thomas P, Jeyarani V, Choephel T, et al. In-vitro cell line Models and Assay methods to study the Anti-diabetic Activity. Research Journal of Pharmacy and Technology. 2019; 12(5): 2200-6. DOI: 10.5958/0974-360X.2019.00367.6

6. World Health Organization. Classification of diabetes mellitus. 2019 https://apps.who.int/iris/handle/10665/325182

7. Hu Y, Chen Y. Overview of Type 2 Diabetes Drugs on the Market. 2020; 1–14. http://www.scirp.org/journal/Paperabs.aspx?PaperID=102005

8. Ge R, Huang Y, Liang G, Li X. 11 -Hydroxysteroid Dehydrogenase Type 1 Inhibitors as Promising Therapeutic Drugs for Diabetes : Status and Development. 2010; 412–22. https://doi.org/10.2174/092986710790226147

9. Anagnostis P, Katsiki N, Adamidou F, Athyros VG, Karagiannis A, Kita M, et al. 11beta-Hydroxysteroid dehydrogenase type 1 inhibitors: Novel agents for the treatment of metabolic syndrome and obesity-related disorders? Metabolism. 2013; 62(1): 21–33. Available from: http://dx.doi.org/10.1016/j.metabol.2012.05.002

10. Kotelevtsev Y, Holmes M.C., Burchell A, Houston P.M., Schmoll D, Jamieson P, et al. 11β-Hydroxysteroid dehydrogenase type 1 knockout mice show attenuated glucocorticoid-inducible responses and resist hyperglycemia on obesity or stress. Proc Natl Acad Sci USA. 1997; 94(26): 14924–9. https://doi.org/10.1073/pnas.94.26.14924

11. Patel HM, Noolvi MN, Sharma P. CHEMISTRY Quantitative structure – activity relationship ( QSAR ) studies as strategic approach in Drug Discovery. 2014; 4991–5007. https://doi.org/10.1007/s00044-014-1072-3

12. Neves BJ, Braga RC, Melo-Filho C.C., et al. QSAR-based virtual screening: Advances and applications in drug discovery. Front Pharmacol. 2018; 9: 1–7. doi: 10.3389/fphar.2018.01275

13. Kaur N, Singh K. 3D-QSAR and Molecular Docking Studies of N-(2-Aminophenyl)-Benzamide Derivatives as Inhibitors of HDAC^ sub 2^. Research Journal of Pharmacy and Technology. 2014; 7(7): 760.

14. Rosy JC, Balamurali S, Mary JA, Shenbagarathai R, Sundar K. Generation of 2D-QSAR model for angiogenin inhibitors: A ligand-based approach for cancer drug design. Trends Bioinforma. 2016; 9(1–3): 1–13.

15. Vaishnav Y, Thakur A, Kaur CD, Verma S, Mishra A, Jain SK, Ghode P, et al. QSAR Analysis of some N, N-diphenyl urea derivatives as CCR5 Receptor Antagonist. MeSO. 2018; 15(7.8239): 14. DOI: 10.5958/0974-360X.2018.00697.2

16. Burns JA, Whitesides GM. Feed-forward neural networks in chemistry: mathematical systems for classification and pattern recognition. Chemical Reviews. 1993; 93(8): 2583-601. https://doi.org/10.1021/cr00024a001

17. Poojita K, Fathima F, Ray R, Kumar L, Verma R et al. Atom based 3D QSAR and Fingerprint based 2D QSAR of Novel Molecules as MmpL3 receptor inhibitors for Mycobacterium tuberculosis. Research Journal of Pharmacy and Technology. 2021; 14(12): 6321-9. DOI:10.52711/0974-360X.2021.01093

18. Xu Z, Tice CM, Zhao W, Cacatian S, Ye YJ, Singh SB, et al. Structure-based design and synthesis of 1,3-oxazinan-2-one inhibitors of 11β-hydroxysteroid dehydrogenase type 1. J Med Chem. 2011; 54(17): 6050–62. https://doi.org/10.1021/jm2005354

19. Dubey S, Bhardwaj S, Parbhakaran P, Singh E. et al. In silico Prediction of Pyrazoline Derivatives as Antimalarial agents. Asian Journal of Pharmaceutical Research. 2022; 12(2): 119-24. http://dx.doi.org/10.52711/2231-5691.2022.00018

20. Dessalew N. Sci Pharm QSAR Study on Novel CCR5 Receptor Antagonists : An Insight into the Structural Requirement for the HIV Co Receptor Antagonist Activity. 2008; https://doi.org/10.3797/scipharm.0807-20

21. Singh S, Das S, Pandey A, Paliwal S, Singh R. Quantitative structure activity relationship studies of topoisomerase i inhibitors as potent antibreast cancer agents. J Chem. 2013; 2013. https://doi.org/10.1155/2013/849793

22. Peter SC., Dhanjal JK, Malik V, Radhakrishnan N, Jayakanthan M, Sundar D, et al. Quantitative structure-activity relationship (QSAR): Modeling approaches to biological applications [Internet]. Vols. 1–3, Encyclopedia of Bioinformatics and Computational Biology: ABC of Bioinformatics. 2018. 661–676 p. Available from: http://dx.doi.org/10.1016/B978-0-12-809633-8.20197-0

23. Dessalew N. Sci Pharm QSAR Study on Novel CCR5 Receptor Antagonists : An Insight into the Structural Requirement for the HIV Co Receptor Antagonist Activity. 2008; https://doi.org/10.3797/scipharm.0807-20

24. Baviskar BA, Deore SL, Jadhav AI. 2D and 3D QSAR Studies of Saponin Analogues as Antifungal Agents against Candida albicans. J Young Pharm. 2020; 12(1): 48–54. https://dx.doi.org/10.5530/jyp.2020.12.10

25. Madhawai K, Rishipathak D, Chhajed S, Kshirsagar S. et al. Predicting the Anti-Inflammatory Activity of Novel 5-Phenylsulfamoyl-2-(2-Nitroxy)(Acetoxy) Benzoic acid derivatives using 2D and 3D-QSAR (kNN-MFA) Analysis. Asian J. Res. Pharm. Sci. 2017; 7(4): 227-34. DOI: 10.5958/2231-5659.2017.00036.4

26. Sainy N, Dubey N, Sharma R, Dubey N, Sainy J. et al. 3D QSAR Analysis of Flavones as Antidiabetic agents. Research Journal of Pharmacy and Technology. 2022; 15(4): 1689-95. http://dx.doi.org/10.52711/0974-360X.2022.00283

27. Prajapati PM, Shah YR, Sen DJ. Artificial Neural Network: A New Approach for QSAR Study. Research Journal of Science and Technology. 2011; 3(1): 17-24.

28. Malik JK, Soni H, Singhai AK. QSAR-Application in Drug Design. International Journal of Pharmaceutical Research and Allied Sciences. 2013; 2(1).

29. Harper KC, Bess EN, Sigman MS, of asymmetric catalytic reactions. Nat Chem. 2012; 4(5): 366–74. Available from: http://dx.doi.org/10.1038/nchem.1297

30. Dessalew N. QSAR study on dual SET and NET reuptake inhibitors: An insight into the structural requirement for antidepressant activity. J Enzyme Inhib Med Chem. 2009; 24(1): 262–71.

Received on 12.10.2022 Modified on 06.02.2023

Research J. Pharm. and Tech 2024; 17(1):347-357.

DOI: 10.52711/0974-360X.2024.00054