Figure 5

Performance comparison of PPI inference methods. Performance of our sampling approach applied to PPI inference methods that operate on binary bait-prey interaction data (Hart et al. [16], PE [9], and SAI [21]), and compared to state-of-the-art methods that make use of spectral counts (HGSCore [6] and SAINT [13]). For each method that operates on binary data, two curves are plotted: (i) a dashed curve that shows the performance of the method when applied to a direct binarization of the spectral count data (i.e., converting all nonzero spectral counts to 1s)—a common approach—and (ii) a solid curve showing performance upon applying our sampling approach with p = 0.3. We evaluate performance according to the number of PPI inferences (out of the highest-confidence 25,000 or 2,500) validated on gold standard tests, as explained in the main text. The plot shows performance relative to a baseline method of simply ranking PPIs in decreasing order of observed spectral counts. All methods were run using default parameter settings.