Abstract
Correlation analysis and, in particular, artificial neural networks (ANN) were used to predict the anti-mycobacterial activity of substituted 3-phenyl-2H-1,3-benzoxazine-2,4(3H)-diones (PBODs) by quantitative structure — activity relationship (QSAR) calculations. Initially, sixty-four derivatives were synthesised and biologically tested; ten further derivatives were proposed for future synthesis on the basis of the prediction results. The biological activity was originally expressed by minimum inhibitory concentration (MIC) against Mycobacterium tuberculosis; however, its transformed pMIC form was found to be more informative. Theoretical molecular descriptors of several types were selected to establish a primary drug model of the species which was expected to exhibit a substantial anti-mycobacterial effect. Lipophilicity and solubility indices, several basic molecular properties, quantum chemistry quantities as well as 1H and 13C NMR chemical shifts, were employed as the descriptors, enabling a very successful prediction of the pMIC values. The utilisation of in silico variables and simulated NMR data is highly advantageous in the first phase of the drug design, as they permit prediction of the compounds with a high expected activity, minimising the risk of synthesising less active species. The MIC values predicted at less than 4 μmol L−1 for six of the ten compounds suggested for further synthesis are better than the best value for the original set of compounds.