Fig. 4
From: PS-MCL: parallel shotgun coarsened Markov clustering of protein interaction networks
The number of nodes belonging to clusters of specific sizes. With balancing factor b=0, i.e. without considering balance, PS-MCL and MLR-MCL do not find clusters properly. In general, they output too large clusters and even group all the nodes into one cluster for MINT. Using balancing factor b>0, both result in clusters whose sizes are concentrated around a small value. That value is larger in PS-MCL than in MLR-MCL. Observe that MLR-MCL makes a significantly large number of tiny clusters including singletons which are meaningless in clustering. In contrast, our proposed PS-MCL greatly reduces that number: less than 5% compared with the number by MLR-MCL. For all cases, MCL suffers from the cluster fragmentation problem. For the other datasets in 3, we observe the same patterns described here