Etude des interactions protéine-protéine et protéine-ligand par bio- et chimie-informatique structurale : Identification de petites molécules bio-actives

摘要

In this document, I describe my contribution into 2 complementary aspects of structural bioinformatic : the modelling of the 3D structure of proteins and the modelling of their modulators. The identification and the structural analysis of protein binding sites (by proteins or small molecules) allow the modulation of their biological function by using new synthetic entities. The last may act as useful tools for further in vivo and in vitro physiopathologic studies. Such interactions can be predicted by docking programs. First, I developed an integrated system for the modelling of the 3D structure of proteins by using homology modelling. The server @TOME ((@utomatic Threading Optimisation Modelling & Evaluation) was the first, in France, that allows the whole automatic modelling of protein structures by performing the 4 usual steps: fold recognition, alignment, model building and model evaluation (http://bioserver.cbs.cnrs.fr). @TOME was evaluated during the CASP5 session in 2002 (http://predictioncenter.llnl.gov/). Results indicate that our server performed well over the 67 targets. @TOME was ranked 26th of the 187 registered groups. More importantly, results validated the use of the fully automatic mode of @TOME on cases where the sequence identity between the target and the template structure is more than 30%. During a sabbatical stay at the University of Stony Brook, NY, USA at Ilya Vakser's lab, I developed a database of annotated protein-protein co-crystallized structures. This bound-bound database provides the foundation of a more ambitious system of databases called DOCKGROUND. It is designed to become a comprehensive public environment for developing and validating new methodologies for modelling of protein interactions. Concerning the bound-bound part, programs filter and annotate structures coming from the Biological Unit database. Our data has several options to exclude particular complexes as well as redundancies based on sequence or structural similarities. The database is accessible by the web (http://dockground.bioinformatics.ku.edu) and is regularly updated. Thanks to a NIH grant, the DOCKGROUND project is under expansion with the release of 3 new databases. De novo drug design and virtual screenings constitute a major part of my researches. More particularly, I develop a de novo drug design program called LEA3D. LEA3D is able to optimally combine fragments to generate ideal putative ligands. These methods have been applied successfully to the Thymidine Monophosphate Kinase (TMPK) of Mycobacterium tuberculosis. In collaboration with chemists and biochemists of Pasteur Institute, new structural families of inhibitors have been identified comprising one synthetic inhibitor with a 3-fold better affinity than the substrate dTMP. A new hit ‘hunting' strategy based on fragment screenings is also described. It combines cheminformatic, NMR and crystallography experimental methods. This strategy aims to identify low molecular weight compounds that are optimized into more elaborated and potent binders.