Figure 4

Distribution of clusters from the COMBINED strategy among any single or multiple number of clustering algorithms that generated them using the SWC network, and their precision (proportion of clusters that match test complexes), in (a) yeast, (b) human. Different clustering algorithms produce different sets of clusters: in either yeast or human, about 85% of clusters are generated by a single unique algorithm, while less than 7% of clusters are generated by three or more algorithms. Thus aggregating clusters from different algorithms increases the recall of complex prediction. Furthermore, precision increases as clusters are generated by a greater number of algorithms: the highest precision of clusters generated by a single algorithm is 16%, increasing to 78% for clusters generated by all algorithms in yeast, and 37% in human.