Skip to main content
Figure 2 | BMC Bioinformatics

Figure 2

From: Markov clustering versus affinity propagation for the partitioning of protein interaction graphs

Figure 2

Effect of random addition of edges to the original unweighted protein interaction graph on the performance of the MCL and AP algorithms. Performance is evaluated by plotting two parameters, the Geometric Accuracy (Acc) (a), and Separation (Sep) (b). The random addition of varying proportions of edges mimics noise created due to varying proportions of False Positive interactions (spurious interactions). For AP, only those points where the algorithm converged are plotted. Definitions of Acc and Sep are given in Methods. The open circle marks the Acc and Sep values achieved by the curated complexes used to generate the original protein interaction graph, as measured against themselves – note that separation is < 1 due to shared components between complexes. Dashed lines indicate the values obtained from random graphs used as controls (see Methods). The solid horizontal line shows the Acc (a) or Sep (b) values achieved by not grouping any proteins (i.e. a "clustering" that consists entirely of singletons).

Back to article page