Fig. 4.

Hamming distance and distribution of module sizes for SNR = 0.3 and sample size 100, generated with randomly signed edges in the graph. bFMD is our balance-based module discovery method. W is the method from Miecznikowski et al. (2016), WGCNA is an implementation of WGCNA using k-means for 6 clusters (Langfelder and Horvath, 2008) and SPC is sparse principal components (Bair et al., 2006). The module sizes are only shown up to 100. The interquartile ranges of module size for each method: bFMD (19, 28), W (149, 258), WGCNA (178, 192) and SPC (37, 108). Since the minimum module size for WGCNA is 155, its distribution is out of the range of this plot