Below are a few notes regarding the contributions I found most relevant to my research interests. Although most speakers gave their talk in French[1], most of the proceedings are available in English.

Simon Fraser (EMBL Heidelberg), Signal and noise in quantitative RNA sequencing (invited talk)
Simon pointed out that, quite absurdly, a lot of RNA-Seq studies published in high-profile places draw conclusions from Fisher tests run on a single sample per tissue / condition. He then presented the DESeq2 package for differential gene expression analysis. DESeq2 performs a shrinkage estimation of dispersion based on generalized linear models and a ridge penalty term.

Alia Dehman (Génopole, Evry), Incorporating linkage disequilibrium blocks in genome-wide association studies (with C. Ambroise and P. Neuvial)
The idea here is to group SNPs that belong to the same LD block together, so that they are selected together (or not at all) by a Lasso type of approach to SNP selection. Alia first presented a three-step approach:

  1. Perform hierarchical clustering of the SNPs based on a measure of LD (D or R²) and Ward's minimum variance criterion;
  2. Infer LD blocks using a criterion such as maximal gap or BIC to choose the level at which to stop the hierarchical clustering;
  3. Use the inferred LD blocks as feature groups in a group lasso approach.

She then suggested an alternative to the threshold/groups determination, by using a tree-guided group Lasso approach directly on the clustering tree.

Magali Michaut (NKI, Amsterdam), Pathway in mutation status predict chemotherapy response in triple negative breast cancer (with E. H. Lips, L. Mulder, M. Hoogstraat, M. H. Koudjis, R. Bernards, J. Wesseling, S. Rodenhuis and L. Wessels)
What interested me most in that talk was the simple but effective core idea of the study: as no single genes were found to be associated with the trait of interest (response to treatment), Magali merged the mutations by pathways, considering a pathway as mutated if at least one of its genes was mutated, and was able to detect enrichment in mutations in one particular pathway in non-responders compared to responders.

Alia's talk, Magali's talk and mine were all part of a parallel session on association studies and variants analysis, and I find it telling that all of our talks revolved around ways of grouping variants or genes together to increase power of detection.

Jean-Loup Faulon (Génopole, Evry), A rational metabolic engineering pipeline: from CAD to product identification (invited talk)
Metabolic engineering is an emerging field, the USDA estimating that it will cover 28% of the chemical market (614 billions USD) by 2025. Currently more than 100 compounds are produced by metabolic engineering; however in most cases it is unclear why these particular genes or pathways have been chosen. This justifies the need for a computer assisted design pipeline such as BioRetroSynth, which includes chemical characterization by ECFP signatures, minimal pathway search, QSAR and flux balance analysis.

Christine Brun (Aix-Marseille Université), The predictive power of protein interaction networks (invited talk)
Protein function is a complex notion, encapsulating many concepts from physiology to development and cellular or function. How can protein-protein interaction networks help elucidate protein function? Networks can be leveraged at either global or local level (or both). Christine then gave examples of such analyses:

  • Determining functional modules. Functional modules are defined as "discrete entities whose function is separable from those of other modules" (Hartwell et al., 1999). The typical approach is to partition (cluster) the network based on various criteria, in particular Newman's modularity. However it makes more sense, biologically, that proteins belong to several modules. This justifies the development of an overlapping cluster generator (Becker et al., 2012). The idea here is to cover the graph with an initial overlapping class system (e.g.: the set of all edges, the set of maximal cliques, the set of centered cliques), then optimize Newman's modularity (Angelleli & Reboul, 2007). This is implemented in a Cytoscape plugin called Clust&See (Spinelli et al., 2013).
  • Deciphering the Painless mechanotransduction pathway: a local analysis. Painless is a Drosophila pathway involved in response to mechanical stress to the larvae. Looking at nearest neighbors of the pathway in a two-hybrid screen interaction network made it possible to understand why the larva's heart can stop when the larva is under mechanical stress.
  • Understanding the role of Hsp27: a local and global analysis. Hsp27 is overexpressed in many cancers, in particular prostate cancer. Inhibition by antisense oligonucleotide OGX-427 (currently under PhaseII trial in North America) induces tumor regression. The goal of that project is to identify other functions of Hsp27 to anticipate potential side effects. The study (to be published soon) built an Hsp27-containing functional modules map. It suggested that Hsp27 is involved in splicing and DNA-repair, was confirmed by experimental assays on prostate cancer cells.
  • Looking for moonlighting proteins: a global analysis. Moonlighting is multi-functionality, at the cellular function level, related to different molecular entities (as opposed to functional promiscuity). Networks can be used to help identify them; there seem to be more of them than first thought.

Guillaume Collet (INRIA Rennes), MINIA on Raspberry Pi: assembling a 100Mbp genome on a credit card sized computer (poster, with G. Rizk, R. Chikhi and D. Lavenier)
The title says it all: genome assembly on a Raspberry Pi.


[1] This was the topic of much discussion and controversy, but I gave mine in English for the following reasons: (1) Some of the people in the audience, including invited speakers, do not speak French fluently or at all; (2) When I attended the German Conference in Bioinformatics in 2011, everything was in English (and I was very grateful for it, although I could probably follow a talk in German, particularly accompanied with English slides); (3) I am so used to discussing research in English, it is actually more comfortable for me than in French, and I therefore think I give better presentations in English than in French. I do recognize JOBIM announces itself as the "annual meeting of the french-speaking bioinformatics community" on their website.