Analysis of Regulatory and Interaction Networks from Clusters of Co-expressed Genes - Clustering Challenges in Biological Network - page 65

Fig. 3.7. A power law distribution of transcription factors amongst a randomly selected group of

genes.

imental data. However, the question of which of these binding sites are active
under a given experimental procedure in complex mammalian systems has not
been adequately answered by either computational or experimental means.
From the analysis, our contention is that the primary biological aspect that
drives transcriptional networks to be scale free is the distribution of transcription
factors by their promiscuity rather than their respective promoter regions. If it was
driven by properties of their respective promoter regions, we would expect a dif-
ferent character in the overall all connectivity property when comparing random
genes, and two different iterations of phylogenetic footprinting. We believe this
to be a significant outcome because it suggests that the creation and evolutionary
conservation of transcriptional links is not driven by mutations in the promoter
region, but rather by mutations in the coding regions of the transcription factors
themselves. While this does not preclude mutations in the promoter region playing
a significant role in genetic diseases or differences in phenotype within a species
by changing the binding affinities of transcription factors, we feel that the evidence
points to the hypothesis that the specificity of a transcription factor mediated by
its amino acid sequence is the primary determinant of the transcriptional network
structure.
It has been shown that the distribution of promiscuous to selective transcrip-
tion factors follows an exponential distribution thereby allowing one to account
for the scale free nature of biological networks. However, there are slight discrep-
ancies in the sub-networks as evidenced by the presence of transcription factors
that are highly conserved amongst the genes within a cluster despite their selec-
tive binding sites. Other transcription factor binding matrices such as PAX4 and
USF1, are much more selective in their binding with 8-10 exact matches in their
recognition sequence but appear as often as the more promiscuous transcription