Structure-based peptide design
Equipe 1 du Dr P. Tufféry (DR Inserm)
Recent progress in synthesis, bioavailability, systemic stability, and selective cell penetration of peptide make them a promising alternative to supplement small compounds as candidate therapeutics, which motivates studies to understand molecular mechanisms underlying their biological functions. Peptides can be endogeneous (hormons, neuropeptides, ...) or exogeneous (AMPs, toxins, ...). "Short Linear Motifs" - SLiMs - that correspond to conserved protein fragments located in disordered regions of proteins, but also more generally protein fragments such as for instance those used in the context of vaccine design can also by extension be assimilated to peptides. Peptide biological functions involve multiple interactions with membranes, nucleic acids and proteins. Particularly, peptide-protein interactions are involved in the regulation of cell and tissue activity, and in the immune system.
Our research is about the in silico characterization of peptides and their interactions. Several directions are considered:
Thème 1 : Peptide and protein fragment de novo 3D structure prediction.
Thème 2 : Peptide-protéine interactions .
Thème 3 : Interaction specificity and similarity.
Thème 4 : Structural Bioinformatics.
In addition, the team hosts the RPBS (Ressource Parisienne en Bioinformatique Structurale) platform. In this context, we also have a more classical structural bioinformatics activity, about the analysis and modeling of structure and function of proteins. On-line tools are available through our Mobyle portal.
The knowledge of peptide 3D structure is a prerequisite for an accurate caracterization of their funcitonal interactions. When estimates of the number of peptide sequences occuring in life are of several millions, only ~ 2 000 structures of peptides between 10 and 50 amino acids were know early 2014.
We have developed PEP-FOLD, an original and accurate approach for the prediction of peptide structure from sequence. PEP-FOLD, still in progress, is able to predict the structure of peptides from 9 up to 50 amino acids, linear or with disulfide bonds.
The detailled caracterization of peptide-protein interactions relies on a 2 step protocol, with (i) the identification of the binding site on protein surface, and (ii) peptide docking to get an accurate conformation of the complex. Our recent efforts have led to the development of PEP-SiteFinder, an approach to predict peptide binding site given the sequence of the peptide and the structure of a protein.
Conformational similarities are a key to assist protein structure modeling (similar folds), but also to caracerize the specificity of interactions, searching for non linear (patch) similarities. The choice of the criterion to measure similarity largely conditions the quality of the results. Commonly used criteria have important limitations such as the dependence on the dimension of the vectors to compare, that introduces fuzziness in the signal. We have recently proposed a new criterion to measure 3D similarity based on a Binet-Cauchy kernel. It is a measure of the geometrical correlation between structures . We are working on the application of such criterion to structural alignment, and to the general search of similarities among collections of un-aligned coordinates.