Fig. 2

Relative performance of taxonomic classifications assigned by each pipeline. A Precision-recall graphs of MetaPro, HUMAnN3, and SAMSA2’s gene annotation strengths against the gold standard hits. True positives are annotations of reads from the pipeline that agree with the gold standard. True negatives are reads that are unidentified by both the pipeline and the gold standard. False negatives are reads that the gold-standard identified but were not annotated by the pipeline. False positives are when the pipeline annotated a read where the gold standard did not. Due to the low number of true positives across all samples and all tools, precision and recall are low, and the plots do not resemble a typical precision-recall plot. B Chord diagrams mapping the relationship of reads annotated by the gold-standard and MetaPro for NOD mouse gut and kimchi datasets. Arcs represent categories of reads. Bands between arcs indicate the proportion of reads mapping between categories, for example while many of reads mapped to Parabacteroides ASF519 are annotated as Parabacteroidesgoldsteinii in MetaPro, other reads are assigned by MetaPro to derive from a variety of other organisms. For the kimchi dataset, we observe a higher agreement between MetaPro and gold standard annotations. C Pie charts showing the breakdown of taxonomic assignments for each pipeline for a selection of samples. Each chart is constructed only from reads deemed to be of putative bacterial mRNA origin as defined by each pipeline. MetaPro annotates more putative reads to species than either HUMAnN3 or SAMSA2