| Quantitative structure-activity relationships (QSAR) attempt to explain the biological activity of a compound from its structural properties. For categorical activities (i.e. active vs. inactive), classification models are developed. Classification and QSAR analysis was performed for a series of hydantoin derivatives having anticonvulsant activity in mice and rats. Traditional descriptors were generated to encode two-dimensional (2D) and simple 3D properties of molecules, and models were developed using descriptor subsets selected with a genetic algorithm. Field-based 3D QSAR models were developed for a series of dihydrofolate reductase inhibitors using the steric, electrostatic and hydrophobic properties of molecules.; A large number of methods are available for QSAR modelling. Descriptors calculated with CoMFA (Comparative Molecular Field Analysis), CoMSIA (Comparative Molecular Similarity Indices Analysis), EVA (Eigenvalue Analysis), HQSAR (Hologram QSAR), and traditional 2D descriptors were used for developing models from 8 datasets. It was found that HQSAR (a method that uses 2D properties) generally performs as well as CoMFA and CoMSIA (methods that use 3D properties); other descriptor sets performed less well. When using traditional descriptors, only neural network ensembles were found to be similarly or more predictive than partial least-squares for fitting models.; The large amount of qualitative data produced by high-throughput screening has spurred the development of classification methods. We have developed and implemented the SFGA method, which fits descriptor splines to activities, selecting descriptors with a genetic algorithm. The method compares favourably to established techniques, with SFGA producing more predictive models for 4 of 5 datasets. Another approach for predicting activities employs receptor surface models (RSMs). The RSM consists of a surface that envelops a set of known actives, aligned using their common features. When used for database searching, a RSM is over-constraining as it restricts access to regions that could be occupied by ligands, such as the solvent-protein interface or unexplored pockets. We described a protocol for developing pruned RSMs using information gleaned from 3D QSAR models; the pruned RSMs retrieve on average 1.8 times more actives than un-pruned RSMs in chemical database searches. |
