46 DEGs were excluded since their expression values were zero in all tissues. Gene pairs with |R| ≥ 0.9 were defined as significantly co-expressed and thus “connected” in subsequent co-expression construction. Network connectedness was measured by the number of connections, i.e., edges [65]. The value of 0.9 was used because the resulting connections exhibited an expected power law feature after testing the choice between 0 and 1 [66]. To generate null distribution of connectedness, we performed 10,000 iterations of the same analysis on the same number of genes that were randomly picked to have their FPKM distribution similar to that of DEGs in each region and stage. More specifically, we ranked genes within each data group according to their expression levels, and then randomly selected one gene whose rank was within 5 of each DEG, to ensure that we avoid potential bias due to differences in expression as genes of lower expression are less likely to form connections with others. Afterwards, we tested if connectedness of DEGs was significantly deviated (one-tailed, larger) from the null hypothesis by simulation. A multiple comparison correction was used to correct p-values [59].Table 1Number of tissues by brain region and developmental stage used in network
