Bayesian mixture analysis for metagenomic community profiling.

Sofia Morfopoulou; Vincent Plagnol

Similarity based methods, using homology search algorithms such as BLAST (Altschul et al., 1990) , are considered the most sensitive methods for read classification (Brady and Salzberg, 2009 ). One of the most popular tools using the output of a similarity search algorithm is MEGAN (Huson et al., 2007) . MEGAN addresses ambiguous matches by assigning reads that have multiple possible assignments to several species, to the taxonomic group containing all these species, or else their lowest common ancestor (LCA). This approach is accurate on a higher taxonomic level. However, it is lacking a formal solution to resolving ambiguous matches. A weakness of the similarity based methods is that a long tail of species, each supported only by a few reads can appear in the results. This results from the classification being decided one read at a time, in contrast to considering all reads simultaneously. Hybrid methods combining composition and similarity information such as PhymmBL (Brady and Salzberg, 2009) and RITA (MacDonald et al., 2012) also tackle one read at a time.