Improve the output of primary screening of biologically active compounds using computational models

The aim of the current project is the development and experimental validation of new chemoinformatcs tools that will help to increase success rate of primary biological screening taking into account molecular activity, side activity and metabolic stability. We follow four main directions of the Project: • Development of the tools for unsupervised selection of diverse libraries of compounds for general purpose high-throughput screening based on data set visualization using advanced dimensionality reduction techniques and 2D molecular representation, and the one based on representation of molecules using 3D pharmacophores, • Development of the tools for creation of focused sets of compounds potentially active against a particular target using two strategies: - selection of compounds from available libraries based on 3D ligand- and structure-based pharmacophores, - generation of virtual focused libraries by enumeration of fragment replacements of a given seed compound(s) leading to a desired potency change. Approach will use Matched Molecular pairs for enumeration of possible fragment replacements and machine learning techniques for potency change prediction, • Development of tools to predict compounds with undesired pharmacological profile (high promiscuity, off-target activity) and chemically unstable compounds to exclude them from screening, • Validation of developed approaches in retrospective studies and their application to real drug design projects devoted to development of MARK4 inhibitors, ligands of cannabinoid CB1 receptor and ligands of adenosine receptors.