metaMatch: un algorithme pour l'assignation taxonomique en métagénomique

摘要

Community ecology faces a new challenge as the next-generation sequencing approaches can yield data from hundreds of microbial community samples. This way, combined with accurate and reliable taxonomic assessment, yields hundreds of new data that will contribute to a better understanding of community assemblies formed under various environmental and historical conditions. Algorithms classifying sequences by comparison to a reference library are the most widely used tools for assessing community composition of environmental samples. However, as they are computationally intensive, almost all these algorithms (most standard being BLAST and similar offsprings) use heuristics designed to speed up the database exploration phase, at the cost of being less strict with the quality of the match between a query and a reference. This problem is naturally distributable, as all comparisons (query, reference) are independent. Here, we present a tool enabling comparisons between queries ( say, one million reads) and reference sequences (say, several thousands), and its implementation on two infrastructures: a cluster in MCIA (Mésocentre de Calcul Intensif en Aquitaine) and a production grid EGI. We show how tracking the large number of jobs generated was nearly impossible with gLite, and how this problem could be solved using Dirac. We compare time and quality between a run on Avakas and on the grid EGI. As a perspective, we will develop a user friendly interface enabling this tool to be used routinely on the grid as a diagnostic for a user not acquainted with computing subtleties of the grid.