Fig. 4

Comparison of TARA-TS and TARA in terms of their alignment and prediction overlaps. Comparison of the selected TARA-TS version and TARA for GO term rarity threshold 50, ground truth dataset atleast1-EXP, and the 90% training test, in terms of overlap between their: a aligned yeast-human protein pairs and b predicted protein-GO term associations. In panel (b), precision and recall are shown for each of the three prediction sets captured by the Venn diagram; TARA++’s predictions are those in the overlap. The overlaps are for one of the 10 balanced datasets; so, the alignment size and prediction number of a method may differ from those in Fig. 3b, where the statistics are averaged over all balanced datasets. Results for the other ground truth-rarity datasets are shown in Additional file 1: Figs. S9–S10