Best methods for taxonomic assignment from shotgun species

This is turning into a frequently-asked-question on Twitter, so here is my $0.02 for the best methods to try:

• Metaphlan2 – relies on mapping to taxon-defining genes. Works very well for well-characterised environments and species, e.g. human gut and pathogens. Search database is small so it is quite rapid. High specificity, potentially low sensitivity in poorly-characterised environments, e.g. soil. https://bitbucket.org/biobakery/metaphlan2

• Kraken – http://ccb.jhu.edu/software/kraken/, even faster than Metaphlan2. Lowest common ancestor (LCA) approach combined with k-mer matching, hence being rapid. As it is k-mer based may suffer from lower specificity, particularly for reference genomes containing erroneous k-mers, and therefore recommend filtering the results. Also see http://www.onecodex.com for a web-based, ultra-fast version.

• DIAMOND combined with MEGAN for LCA – DIAMOND is Daniel Huson’s fast replacement for BLASTX. Other alternatives include RAPSearch2. Searching against Genbank non-redundant proteins (nr) is probably the highest sensitivity method for metagenomics assignments but it needs to be combined with an LCA approach to give appropriate specificity (if you don’t believe me, try taking E. coli and shredding it into 100-mers and BLASTXing it and see how many taxa you retrieve). http://ab.inf.uni-tuebingen.de/software/diamond/ http://ab.inf.uni-tuebingen.de/software/megan5/

• If you are boldly exploring strange new worlds to seek out new life, a phylogenetic approach may be more suitable. These tend to be slow, and rely on the presence of one or more marker genes in your sample, but have the advantage of giving a feel for the phylogenetic position of species in your sample. If you are doing this kind of thing I recommend trying Phylosift http://phylosift.wordpress.com/ or MOCAT http://vm-lux.embl.de/~kultima/MOCAT/