Skip to main content
. 2021 Jan 29;22:34. doi: 10.1186/s12859-021-03971-6

Fig. 3.

Fig. 3

Comparison of the three TARA-TS versions and TARA. Comparison of the three TARA-TS versions and TARA for GO term rarity threshold 25 and ground truth dataset atleast1-EXP, in terms of: a classification accuracy, b protein functional prediction accuracy, ) overlap between aligned yeast-human protein pairs, and d overlap between predicted protein-GO term associations. In panel (b), the alignment for e.g., TARA contains 1,716 aligned protein pairs and predicts 3474 protein-GO term associations. In panels (c)–(d), the pairwise overlaps are measured via the Jaccard index. Panel (a) encompasses all y percent training tests. Panels (b)–(d) are for the 90% training test. Comparisons of different metapath choices for metapath2vec can be found in Additional file 1: Fig. S1. Results for the other ground-truth rarity datasets and percent training tests are shown in Additional file 1: Figs. S2–S8