only one of the compounds inter acted with less than five kinases. Many promising kinase inhibitors were abandoned early due to toxicity. EPZ-5676 FDA Yet another common reason for failure was lack of clinical efficacy. The latter problem can be attributed to the multitude and complexity of cellular signaling cascades, with redundant pathways and com plex feed back mechanisms. Use of multi targeted com pounds that can selectively inhibit a specific group of kinases of such pathways might increase the chance to achieve clinical antitumor activity. Yet another reason for lack of clinical efficacy is resistance that arises due to mutations in the targeted oncogene. E. g, drug resistance in imatinib treated leukemia patients appears due to mutations in the BCR ABL fusion protein.

This prompts the need for new generations of drugs that can override the acquired resistance by inhibiting the mutated onco Inhibitors,Modulators,Libraries gene. A computational method widely applied in drug design is quantitative structure activity relationship modelling. QSAR models are used to optimize lead com pounds for target activity and other properties and to perform virtual screening to find new hits. However, drawbacks of QSAR are that its models consider only properties of ligands and that it analyzes interactions with only one drug target at Inhibitors,Modulators,Libraries a time. Hence QSAR models are unable to generalize between multiple targets. A more general approach is proteochemometric mod elling, which we introduced some time ago to study dif ferences in mechanisms of molecular recognition for groups of related proteins.

Inhibitors,Modulators,Libraries Proteochemometric models are based on experimentally determined interac tion data for series of proteins interacting with series Inhibitors,Modulators,Libraries of ligands, like organic compounds, peptide inhibitors, sub strates, etc. These data are correlated to descriptors of the two sets of interacting entities, which creates models that can be used to predict activities of yet untested ligand protein combinations, as well as foresee activity profiles of novel unseen ligands and proteins. Proteochemometric models take advantage of the fact that 3 D structures of homologous proteins are more con served than their primary sequences and functions. Thus, proteins that Inhibitors,Modulators,Libraries have diverged functionally during evolution may still share the same structural organization and exploit similar molecular interaction mechanisms. The principle behind proteochemometrics is simple. It requires consistent interaction data, numerical descriptions of relevant physico chemical and or struc tural properties of both ligands FTY720 and the protein macro molecules, and a non linear correlation method that jointly uses the two sets of descriptors to explain ligand protein complementarities and interaction profiles.