Fig. 1

Metagenome phyletic profiling provides novel high-confidence gene function predictions. a Distribution of MPP-H and MPP-O model accuracies (expressed as AUPRC score) on 451 and 325 learnable GO functions, respectively. GO functions are stratified according to their frequency of occurrence, expressed as information content (IC). Baseline classifiers are constructed from randomized MPP-H and MPP-O data obtained by randomly reassigning GO functions to COG/NOG gene families. b Precision-recall curves for example GO terms learnable only by MPP-H (“methanogenesis”) and only by MPP-O (“photosynthetic membrane”). c Predictive accuracies (as AUPRC) for photosynthesis-related gene functions of MPP-O models using deep sea samples and MPP-O models using euphotic zone samples. d Overlap between MPP-H and MPP-O inferences, expressed as percent of GO functions predicted only by MPP-H, only by MPP-O or by both. e Overlap between MPP-H and MPP-O in terms of particular COGs/NOGs to which only one or both types of MPPs can assign those GO terms that could overallbe predicted by both MPP types (green part of the histogram in (d)). f–h Overlap between different instances of MPP and their matched PPs (Methods), expressed as percent of GO functions predicted only by MPP, only by PP or by both. MPP, metagenome phyletic profiles; PP, phyletic profiles; AUPRC, area under the precision-recall curve