Analysis of Regulatory and Interaction Networks from Clusters of Co-expressed Genes - Clustering Challenges in Biological Network - page 66

Table 3.1. Transcription factors conserved more than 95% of the time between mouse and rat. In
contrast to the human/rat case, all of the clusters show transcription factors conserved more than 95%
of the time as well transcription factors which are highly conserved and not found in a random sampling
of genes.
factor like STAT5 or STAT6. There is the question of whether the deviations
from the power-law distribution due to these transcription factors is a significant
characteristic of transcriptional network, or whether these deviations are within
acceptable deviations from the parameterized exponential curve. However, given
that these links are not present during a random sampling of promoter regions,
we believe that these links represent additional information provided by the gene
selection and classification step. Given that even after clustering, the transcription
factors in Table 3.1 are still relatively few, they could have been missed in a global
assessment of network properties, but still may have an important effect upon the
robustness of the network. These highly connected nodes which show up only in a
set of highly correlated genes may serve to function as a means for preserving the
connectivity structure of a network given the removal of a highly connected hub.
Recent experimental evidence [61] has suggested that in fact the removal of highly
connected hubs is not as lethal in biological organisms as it is to purely scale free
networks such as the Internet. Our computational assessment seems to agree with
this assessment, and it suggests that the process of clustering has allowed for the
possible identification of more hubs than would be suggested under a purely scale
free topography.
Another interesting observation is that while the whole transcriptional network
appears to be scale-free in nature, each of the small co-expressed subsets also ap-
pears to be part of individual scale-free sub-networks. It has been shown that if
in network construction, newer pathways had starting points in pre-existing path-
ways much like the way new hyperlinks on the Internet tend to link to older more