Abstract

That the implementation of Electronic-Topological Method and a variant of Feed Forward Neural Network (FFNN) called as the Associative Neural Network are applied to the compounds of Hydrazones derivatives have been employed in order to construct model which can be used in the prediction of antituberculosis activity. The supervised learning has been performed using (ASNN) and categorized correctly 84.4% of them, namely, 38 out of 45. Ph1 pharmacophore and Ph2 pharmacophore consisting of 6 and 7 atoms, respectively were found. Anti-pharmacophore features socalled “break of activity” have also been revealed, which means that APh1 is found in 22 inactive molecules. Statistical analyses have been carried out by using the descriptors, such as E_HOMO, E_LUMO, ΔE, hardness, softness, chemical potential, electrophilicity index, exact polarizibility, total of electronic and zero point energies, dipole moment as independent variables in order to account for the dependent variable called inhibition efficiency. Observing several complexities, namely, linearity, nonlinearity and multi-co linearity at the same time leads data to be modeled using two different techniques called multiple regression and Artificial Neural Networks (ANNs) after computing correlations among descriptors in order to compute QSAR. Computations resulting in determining some compounds with relatively high values of inhibition are presented.

Keywords: Antimycobacterial activity, associative neural network, DFT, electronic topological method, hydrazide-hydrazones, QSAR.

Graphical Abstract