Section: Research Program

Axis 2: Sequence comparison

Comparing genomic sequences (DNA, RNA, protein) is a basic bioinformatics task. Powerful heuristics (such as the seed-extend heuristic used in the well-known BLAST software) have been proposed to limit the computation time. The underlying data structures are based on seed indexes allowing a drastic reduction of the search space. However, due to the increasing flux of genomic sequences, this treatment tends to increase and becomes a critical section, especialy in metagenomic projects where hundred of millions of reads must be compared to large genomic banks for taxonomic of functional assignation.

Our research follows mainly two directions. The first one revisits the seed-extend heuristic in the context of the bank-to-bank comparison problem. It requires new data structures to better classify the genomic information, and new algorithmic methods to navigate through this mass of data [7] , [10] . The second one addresses metagenomic challenges that have to extract relevant knowledge from Tera bytes of data. In that case, the notion of sequence similarity itself is redefined in order to work on objects that are much simpler than the standard alignment score, and that are better suited for large-scale computation. Raw information (reads) is first reduced to k-mers from which high speed and parallel algorithms compute approximate similarities based on a well defined statistical model [4] .