Section: New Results

NGS for genotypic variation detection

The computational work on NGS data concerned both algorithmic design and complexity analysis.

Based on the idea that each genotypic variation will correspond to a recognisable pattern in a de Bruijn graph constructed from a set of sequence reads, we had proposed a generic model for SNPs in DNA data, and then generalised it to the analysis of RNA. In this case, not only SNPs are present but also alternative splicing (AS) events, which, once again, generate a recognisable pattern in the de Bruijn Graph. We had therefore proposed a general model for all these variations (SNPs, indels and AS events) and introduced an exact algorithm (KisSplice ) to extract all alternative splicing events. The algorithm also outputs candidate SNPs and indels. This year, we improved the algorithm [26] . As the problem relates to an old one in algorithmics (cycle enumeration), we also revisited it from a theoretical point of view [23] .

The improved version of KisSplice [26] was used to analyse RNAseq data from two lines of Asobara tabida exhibiting different ovarian phenotypes in the absence of its endosymbiont Wolbachia. Although infected individuals of the two lines have similar phenotypes, numerous genes are differentially expressed between the two infected conditions. This could mean that two divergent strategies of tolerance have evolved. Preliminary results on the analysis of polymorphisms between these two lines suggest that differentially expressed genes tend to accumulate more variation. We are currently, via experiments done by the biologists in our team, testing the hypothesis that such genes are under strong selection pressure and may evolve through mutation accumulation, a process that could be related to assimilation.

A preliminary analysis of human data from the ENCODE project performed with KisSplice showed that an assembly-based method (without reference genome) is able to recover AS events that are missed by mapping-based methods (with a reference genome). Some of these events were experimentally validated, which represents the best type of proof we can provide to the biologists. The experimental part is made by our collaborator from the Inserm, Didier Auboeuf, in his team at the Centre National de Cancérologie of Lyon (CNCL), with whom we had an Inserm project, EXOMIC, funded for three years starting from 2012.

The identification of SNPs is also getting renewed interest even in the presence of a reference genome thanks to the possibility of re-sequencing many times the genome of a same or of very closely related species. The difficulty in the case of SNPs is to distinguish them from sequencing errors and from inexact repeats. We proposed a statistical test enabling to identify variations that are condition-specific, which enables to greatly enrich the list of potential SNP candidates. The paper on this test is in preparation. Its results as applied to the RNAseq data from two lines of Asobara tabida (see above) and to Drosophila species having diverged very recently were validated by, respectively, Fabrice Vavre and Cristina Vieira, both members of BAMBOO.

We also started addressing the problem that repeats (such as transposable elements for instance but not only) represent more in general for both local and global assemblers. We are thus developing a method that would enable to identify, in a de Bruijn graph built from RNAseq data, the vertices potentially corresponding to the borders of a repeated sequence. Preliminary results on simulated and real data show that the approach is promising (paper in preparation).