PRECISION TARGETING OF THE HUMAN KINOME: BIG DATA, BIG LIBRARIES

Any lineup of the most transformative “targeted” cancer therapeutics from the past 20 years is inevitably dominated by kinase inhibitors. In retrospect, this is because collectively, the assemblage of 500 protein kinases in humans controls virtually every pathway that can be dysregulated in cancer; and because individually, kinase active sites are highly amenable to inhibition by drug-like chemical matter. Dedicated efforts from both industry and academia have by now provided a tremendous wealth of extant knowledge: numerous crystal structures demonstrating how assorted kinase inhibitors engage their targets, and huge datasets defining which inhibitors bind to which kinases.

The Karanicolas lab has recently developed a comparative modeling approach that allows us to computationally predict – rapidly and with atomic accuracy – the three-dimensional structure of any given inhibitor/kinase pair. We have now begun extending this research in two new directions. First, we are combining our comparative modeling with cutting-edge machine learning approaches, in order to predict both the binding affinities and the selectivity of any arbitrary candidate kinase inhibitor. Second, we are co-opting and generalizing several select synthetic routes that have been used to produce stereotypical kinase inhibitors: through this “computational combichem” strategy we seek to build in silico kinase inhibitor libraries of unprecedented size. Putting the pieces together, we aim to extract from this library novel compounds that potently and selectively inhibit key cancer-driving kinases: these will serve both as tool compounds for discovering new biology, and as potential starting points for developing new therapeutic agents.