An Improved, Bias-Reduced Probabilistic Functional Gene Network of Baker's Yeast, Saccharomyces cerevisiae
YeastNet v. 2 outperforms YeastNet v. 1 , as measured by precision and recall of either genes or reference linkages on independent reference sets.
(A) shows performance on the MIPS functional linkage reference set, with or without masking the term “protein synthesis”, while (B) shows performance on reference linkages derived from KOG functional categories. For both reference sets, we observe significantly improved recall by YeastNet v. 2 for both yeast genes and reference linkages. The effect of the annotation “protein synthesis” is revealed by a significant drop of precision in YeastNet v. 1 but not v. 2 after masking the term during benchmarking. The higher fraction of linkages derived from “protein synthesis” in KOG (∼17% of total linkages) than MIPS (∼4% of total MIPS linkages) explains the apparently higher precision of v. 1 than v. 2 when including the term in (B), resulting in a correspondingly larger drop in precision of v. 1 when the term is masked. All analyses in (A) and (B) are for the 34,000 most confident linkages of each network.