Structure-activity approaches for prediction of chemical reactivity and pharmacological properties of some heterocyclic compounds

Abstract

Benzodiazepine drugs are widely prescribed to treat many psychiatric and neurologic disorders. As its pharmacological action is exerted in a sensitive area of the brain; ''the central nervous system'', it is crucial to provide detailed reports on the chemistry of benzodiazepines, model the mechanism of action that occurs with GABAA receptors, identify the overlap with other modulators, as well as explore the structural requirements that better potentiate the receptor response to benzodiazepines. This dissertation consists of two original studies that consider the new lines of research related to benzodiazepines, particularly the identification of three new TMD binding sites. The first study provides, on the one hand, an overview of the chemistry of six Benzodiazepine basic rings starting from structural characteristics, electronic properties, Global/local reactivities, up to intermolecular interactions with long-range nucleophilic/electrophilic reactants. This was achieved by combining a DFT investigation with a quantitative MEP analysis on the vdW surface. On the other hand, the performed molecular docking simulations allowed identifying the best binding modes, binding interactions, and binding affinities with residues, which helped to validate the quantitative MEP analysis predictions. The second study was conducted on a dataset of [3H]diazepam derivatives. First, molecular docking simulation was used to initially screen the dataset and select the best ligand/target complexes. Afterwise, the best-docked complexes were refined by performing molecular dynamics simulation for 1000 picoseconds. For both simulations, the binding modes, binding interactions, and binding affinities were thoroughly discussed and compared with each other and with outcomes collected from the literature. Additionally, the good pharmacokinetic properties (ADME prediction) as well as compliance with all druglikeness rules were checked via in silico tools for all the dataset compounds. Finally, a QSAR analysis was carried out using an improved version of PLS regression. Briefly, the dataset is randomly split into 10 000 training and test sets that involve, respectively, 80% and 20% of chemicals. Subsequently, 10 000 statistical simulations were conducted that; after excluding outlying observations, yielded 10 000 best training models following the Bayesian Information Criterion. Among these 10 000 best models, the best predictors with the highest probability of occurrence were selected. As a consequence, the derived PLS regression equation explains 63.2% of the variability in BDZ activity around its mean. Furthermore, Internal and external validation metrics assure the robustness and predictability of the developed model. The developed model was interpreted based on literature investigations and a combination of implemented approaches.