Fig. 2

Gene module stability increases with sample size and module size. For each n, we sampled without replacement from all available gene expression profiles 10 times. In each case, a reference module set was produced (by WCGNA), 100 bootstrap re-samplings of the selected expression profiles generated, and the stability of the reference module set across bootstrap re-samplings determined as described in the Methods section. a Stability of the modules is visualized at n = 10, 20, 40, 80, 120 and 160 for the 1st, 5th and 10th rank modules. b Stability is plotted against module size at n = 10, 20, 40, 80, 120 and 160. The dotted line depicts the best-case stability of random modules in simulation. We compare the stability of S (1^st quartile) and XL (4^th quartile) modules using Wilcoxon’s rank-sum test