Corresponding Author Priya Ojha. QSAR (Quantitative structure activity relationship) is a powerful and mathematical technique to set off the correlation in between chemical structure to their biological activity. It was performed on a series of amideoxadiazole-aniline derivative with activity against DGAT-1 employing various physiochemical parameters like topological, lipophilic and electronic. The best model was generated and shows good correlative and predictive ability with values S = 0.33,

more »

... = 41.91, r = 0.94, r² = 0.88, r² (cv) = 0.84 was developed using stepwise MLR and a comparable PLS and FFNN model with r² (cv) = 0.89, 0.88 and 0.86 respectively. After the data reduction, five promising descriptors left were total dipole moment, Log P, VAMP total energy, VAMP LUMO and VAMP HOMO. In addition of QSAR modeling, Lipinski's rule of five was also employed that check the pharmacokinetic profile of the model. The similarity (CARBO and HODGKIN) analysis was also done on the same series which positively support the previous results. Diabetes mellitus is a chronic metabolic disease characterized by hyperglycemia, hyperlipidemia, hyperaminoacidemia, and hypoinsulinaemia. 1 Type II diabetes is a more common form of diabetes constituting 90% of the diabetic population moreover, it is a polygenic disease that results from a complex interplay between genetic predisposition and environmental factors such as diet, degree of physical activity and age. 2 Triacylglycerol (TG) is a highly efficient energy storage form critical for surviving periods of starvation and extended physical activity. 3 Diacylglycerolacyltransferase (DGAT) enzymes catalyze the formation of an ester linkage between a fatty acyl-CoA and the free hydroxyl group of diacylglycerol this action take place in two pathway Glycerol Phosphate and monoacylglycerol. DGAT possesses two isoforms DGAT-1 and DGAT-2. DGAT-1 catalyses the last step of triacylglyceride biosynthesis, transforming diacylglycerol and acyl-CoA into triglyceides. 4, 5 Inhibiting of DGAT-1 might represent a novel approach for that improvement of insulin sensitivity. 6 There are numerous examples in the literature for the successful use of classical descriptors in QSAR. 7, 8 In the view of this, we decided to developed models from classical QSAR descriptor using MLR, PLS and FFNN method to establish the individual and common structural requirement for effective binding of DGAT-1 antagonist. Material and methods:- Data set and Biological activity: The data set containing 48, amide oxadiazole aniline 9 with anti-diabetic activities (Table 1) were taken for present studies in view of high structural diversity and sufficient variation in biological activities. Experimentally determined IC 50 values (µM) of the derivatives were converted into the negative logarithm (Log IC 50 ). Generation of structure: All the chemical structures (anti-diabetic activity) were sketched with the help of Accelrys (Discovery studio version 2.0) and imported into the worksheet of TSAR 3.3 software as .mol files. 10 Defining substituents and Energy optimized structure building: The series has two major substituent's (R 1 and R 2 ) that were defined using "define substituent" option in the TSAR worksheet toolbar. All the loaded structures and their substituent's were converted into three-dimensional (3D) molecular structures by using Cornia make 3D option and further subjected to optimization using Cosmicoptimize 3D option, which includes valence terms as bond potential, bond angles and non-bonded terms as electrostatic interaction and Vanderwaals interaction. The force field supplied by "Cosmic" for energy calculation during a flexible optimization ensures that only the energetically realistic conformations are considered. 11 Calculation of Descriptors and Data reduction: Initially more than 250 descriptors were calculated for both whole molecule and substituent's separately in TSAR software program. TSAR is an integrated analysis package for the interactive investigation of quantitative structure-activity relationships. It automatically calculates numerical descriptors of molecular structure. The calculated descriptors included molecular attributes, molecular indices, atom count and VAMP parameters. 12 The 48 molecules of the series were randomly divided into training set (32 molecules) and test set (11 molecules). Molecules in a training set further used for multiple linear regressions (MLR), partial least square (PLS) and feed forward neural network (FFNN) model development and test set consisting of 11 molecules which were kept on the other hand for future use to check the predictive power of the development model. There is a significant requirement of data reduction to eliminate the chance of correlation. Correlation matrix was used to reduce the number of descriptors and to identify the best subset of with minimum inter-correlation, than checked the other two descriptors. Pair-wise correlation coefficient was calculated for all the paired descriptors. If the inter-correlation coefficient >0.5 was detected, then the descriptor with high correlation with biological activity was kept and others were discarded. This was done with the intention to get the descriptors which are less correlated to each other (independent in the true sense) and highly correlated to the biological activity. 13,14 Thus, finally five independent molecular descriptors, total dipole moment (subst. 2), log P (whole molecule), VAMP total energy (whole molecule), VAMP LUMO (whole molecule) and VAMP HOMO (whole molecule) were fetched and all the descriptors shows high correlation to the biological activity but did not have any correlation among each other.