Section: New Results

3D Modelling and Interactions

Transmembrane beta-barrel proteins (TMB) account for 20 to 30% of identified proteins in a genome but, due to difficulties with standard experimental techniques, they are only 2% of the RCSB Protein Data Bank. Therefore, we study and design algorithmic solutions addressing the secondary structure, an abstraction of the 3D conformation of a molecule, that only retains the contacts between its residues. Although this representation may disregard some of the fine details of the molecule conformation, it still retains the general architecture of molecules, and is especially useful in the study of RiboNucleic Acids (RNAs) and transmembrane beta-barrel proteins (TMB). The latter class of proteins accounts for 20 to 30% of identified proteins in a genome but, due to difficulties with standard experimental techniques, they constitute only 2As TMB perform many vital functions, the prediction of their structure is a challenge for life sciences, while the small number of known structures prohibits knowledge-based methods for structure prediction. As TMBs are strongly structured objects, model based methodologies [26] , [25] are an interesting alternative to these conventional methods. The efficiently obtained 3D structures provide a good model for further 3D and interaction analyses.

In a recent work [34] , we focused on the identification of protein-protein complexes based on the putative interaction between pairs of proteins as the sole source of information. From the results obtained on E. coli, we started working on the prediction of multi-body protein complexes from sequence information alone.

In our protein-RNA project, we managed to obtain the first learning results. We optimized the RosettaDock scores and showed that such an optimization cannot be done efficiently without expert knowledge. The first results are to be presented at EGC in 2014 [61] .

The year 2013 saw the conclusion of a long-term collaboration, involving A. Carbone (UPMC) and A. Lopes (IGM, Paris XI). In a recent paper published in the prestigious Plos Computational Biology [16] journal, we showed that combining coarse-grain molecular cross-docking simulations and binding site predictions based on evolutionary sequence analysis is a viable route to identify true interacting partners for hundreds of proteins with a variate set of protein structures and interfaces. Also, we realized a large-scale analysis of protein binding promiscuity and provided a numerical characterization of partner competition and level of interaction strength for about 28000 false-partner interactions. Finally, we demonstrated that binding site prediction is useful to discriminate native partners, but also to scale up the approach to thousands of protein interactions. This study was based on a large computational effort made by thousands of internet users helping the World Community Grid over a period of 7 months.