Figure 5
From: High-precision high-coverage functional inference from integrated data sources
Comparison of decision rules. (A): Comparative performance of MR, NW, MW, and SGA in annotation precision-coverage analysis. Six data sets are integrated using a linear SVM to construct a weighted FLN. Given a prespecified linkage weight cutoff, MR, NW, and MW use the highest ranked predicted annotations for comparison, SGA takes all the candidate annotations as predictions without ranking. A coverage of 1 means all 5475 proteins can be assigned at least one function. (B): Comparative performance of MR, NW, and MW in annotation precision-recall analysis in a high-coverage and relatively noisy FLN composed of 5475 proteins. (C): Comparative performance of MR, NW, and MW in annotation precision-recall analysis at a low-coverage and relatively confident FLN composed of 1872 proteins. In all the figures, the same evaluations are repeated in random control networks as dashed lines with error bars denoting one standard deviation of 10 repeating runs.