Pairwise Pearson correlation coefficients (PCC) are calculated and a pair of genes are linked by an edge if the PCC value is larger than 0.5, similar to the method used in [ 3, 26].
Moreover, the PCC value for this reason is also negative.
The PCC value of these 10 stable features is insensitive to noise level variation.
This is demonstrated by the observation that networks built using the PCC measures between randomly permuted gene expression profiles, thus preserving some transitivity, also have values of < C> that are far lower than the observed values: human = 0.0933, mouse = 0.1229.
Eq. (2) insures that at the least one of the PCC values exceed the threshold min PCC, while Eq. (3) requires that the difference between two PCC values in different conditions is big enough, i.e., larger than diff PCC.
Note here that we introduced 'MR' by taking a geometric average of the PCC rank from gene A to gene B and that of gene B to gene A, because the two ranks between genes A and B are usually different and thus are not convenient to use like a PCC value.
For example, hsa-let-7a-1 has no coexpressed gene when using 0.8 PCC value but has 29 coexpressed genes if 0.6 PCC value was applied.
To identify putative target-target interactions, the Pearson Correlation Coefficient (PCC), which is known to provide information about the "shape" of gene expression changes [ 47, 48], was used; an absolute PCC value of 0.3 was used as a cutoff.
Using a 0.6 PCC value as the cutoff, 260 ALSGs (75%) and 250 CBSGs (80%) with microarray probes were regarded as co-expressed with ≥ 1 gene(s) with GO annotations.
On the other hand, when we used PCC rank instead of PCC value, the distribution aligned on the diagonal (Fig. 3A, right panel), meaning there was a linear correspondence of PCC ranks between the different samples.
