Binding site detection via mutual information

About 40% of all marketed drugs have the so-called G-protein coupled receptors (GPCRs) as their target protein. There exist more than 800 different GPCRs in humans, of which at least 300 GPCRs are believed to be druggable. Yet, for only two GPCRs there are three-dimensional (3D) protein crystal structures available and consequently little is known about the molecular interactions between pharmacologically active substances and the receptors in this important drug target protein family. A chemogenomics approach as an alternative to the lack of 3D structural information appears attractive for the rational design of GPCR drugs, as an enormous amount of biological activity data for various GPCRs exist and can be used to deduce GPCR structure-function relationships. In this work, we suggest a new approach for the detection of interdependence of features in the GPCR protein sequences and properties of the related small molecule ligands based on mutual information between ligand and sequence space.