Classification of kinase inhibitors using BCUT descriptors.

BCUTs are an interesting class of molecular descriptor which have been proposed for a number of design and QSAR type tasks. It is important to understand what kind of information any particular descriptor encodes and to be able to relate this to the biological properties of the molecules. In this paper we present studies with BCUTs for the classification of ATP site directed kinase inhibitors active against five different protein kinases: three from the serine/threonine family and two from the tyrosine kinase family. In combination with a chemometric method, PLS discriminant analysis, the BCUTs are able to correctly classify the ligands according to their target. A novel class of kinase inhibitors is correctly predicted as inhibitors of the EGFR tyrosine kinase. Comparison with other descriptor types such as two-dimensional fingerprints and three-dimensional pharmacophore-based descriptors allows us to gain an insight into the level of information contained within the BCUTs.