Applications of genetic algorithms on the structure–activity correlation study of a group of non-nucleoside HIV-1 inhibitors

Genetic algorithms (GAs) have been proven to be very useful in data analysis and can be applied as a very powerful technique in quantitative structure–activity relationship (QSAR) analysis. QSAR based on GAs allows the construction of models competitive with or superior to standard methods; moreover, from the analysis of the calculation results, we may get very useful additional information which cannot be provided by other methods. We developed a QSAR program combining genetic algorithm with multiple linear regression and cross-validation. We use it in the QSAR analysis of 23 HIV-1 inhibitors pyrrolobenzothiazepinones (PBTP) and pyrrolobenzoxazepinones (PBP). A group of suitable QSAR models has been obtained. Using the best model we predicted the RT activities of some compounds whose RT experimental activities are unknown. Moreover, from the statistical analysis of the multiple models, we found that low lipophilicity at C-6, small compounds surface, high π electron density of the benzo fused ring and low dipole along the z axis were the most important factors that may influence the RT activities. These descriptors allow a physical explanation of hydrophobic interaction, electronic and steric effect contributing to HIV-1 inhibitory potency.