Kinase Scan Library

The kinase Scan method is a powerful tool used in drug discovery to evaluate the interaction of small molecules with a broad panel of protein kinases. The commonly used technique involves immobilizing a large number of distinct kinases on a solid surface, such as beads, which are then exposed to a drug candidate. At ChemDiv, we evaluated this technology and implemented this approach for a graph neuronal network-driven prediction approach that allows to select chemically diverse compounds potentially engaging with one or more kinases in the target set. In the creation of this library, we used a set of 50 kinases generally involved in the development of cancer. 

Our Kinase Scan collection comprises thousands of highly selective and potent small molecule inhibitors targeting key kinases implicated in the onset and progression of cancer and other diseases. With 50 kinases tested against nearly 17,000 small molecule inhibitors, the structurally diverse set of "hit" compounds offers a wide array of promising candidates for focused drug discovery projects in specific therapeutic areas with a high potential for clinical success in the future. 

Our software algorithm is implemented to provide higher affinity and selectivity of compounds. It operates by processing an ensemble of pharmacophore models representing various molecular conformations through a graph neural network. Initially, it generates a set of three-dimensional conformations for each molecule. Then, for each conformation, it creates a corresponding three-dimensional pharmacophore hypothesis. Alongside the pharmacophore data, a vector comprising molecular fingerprints and physicochemical descriptors is fed into the neural network. Within the network, each pharmacophore hypothesis is independently processed and transformed into a numerical vector of a fixed length. These vectors are then averaged to form a generalized pharmacophore vector. This vector, combined with the molecular fingerprints and physicochemical descriptors, is conveyed to the network's final layer. The output consists of 50 values, ranging from 0 to 1, indicating the likelihood of the studied molecule affecting each of the 50 targets. For each target, a specific binarization threshold is applied to convert these probabilities into a binary score, where 0 indicates IC50 > 1 µM and 1 indicates IC50 < 1 µM.