Figure 2

Classification accuracy and speed comparison of classification programs for three simulated metagenomes. For each metagenome, genus precision and sensitivity are shown for five classifiers, and speed is shown for five programs (PhymmBL65 is simply a confidence-filtered version of PhymmBL’s results, and MetaPhlAn only classifies a subset of reads that map to one of its marker genes, as it is an abundance estimation program). Results shown are for: (a) the HiSeq metagenome, consisting of HiSeq reads (mean length μ = 92 bp) in equal proportion from ten bacterial sequencing projects; (b) the MiSeq metagenome, consisting of MiSeq reads (μ = 156 bp) in equal proportion from ten bacterial projects; and (c) the simBA-5 metagenome, consisting of simulated 100-bp reads with a high error rate from 1,967 bacterial and archaeal taxa. Note that the horizontal axes in all speed graphs have a logarithmic scale.