Fig. 5

Validation of gene function inferences made by MPP on independent data sets. a Average information accretion, per COG, of the novel gene function annotations assigned by MPP, by PP or by both, versus information per COG in the currently known annotations. b Examples of annotations validated in the E. coli and P. aeruginosa CAFA2 benchmark data. Columns represent precision (Pr) scores assigned by MPP-I, Pr scores assigned by PP and confirmed predictions on CAFA2 for six example genes from either E. coli or P. aeruginosa genes. Rows are GO functions. Red color in MPP-I and PP columns indicates that a classifier did not predict a GO function at Pr ≥ 50%. The complete set of predictions is given in Additional file 6: Table S5. Bottom part of heatmap shows a sample of gene functions that did not receive CAFA2 annotations in the shown gene. c Precision-recall curves represent average data over individual curves for 629 GO terms. VAL, validation; CV, cross-validation. d, e Precision-recall curves for individual GO terms computed from an independent validation set of most recent UniProt-GOA annotations. f Proportions of unannotated COGs that received at least one new Uniprot-GOA validated annotation by MPP-I, by PP or by both