Biology Reference
In-Depth Information
with metabolic pathways are available on the KEGG
68
database.
Although functional annotations are incomplete, and sometimes even
wrong, very small
p
-values usually indicate a functional link for the
module.
3.2. Module Visualization
Standard hierarchical clustering still remains the default analysis tool for
large sets of biological data, despite the limitations of this analysis
method for large-scale data.
7,27-32
One reason for this is that the widely-
used representation of expression data in terms of a reordered color-
coded matrix with dendrograms delineating the clusters and their hierar-
chy has the advantage of being exceedingly simple. In particular, many
biologists apparently appreciate that the original expression values (or
ratios) are shown in this presentation (somewhat akin to the fact that
showing the image of a gel shift experiment is still a must, although
quantification software for gels has existed for some time). Accordingly,
we have designed a new visualization tool, ExpressionView, which pres-
ents modules as rectangles that denote its genes and arrays on the actual
expression data (see Fig. 7). Since it is in general impossible to represent
more than two mutually overlapping modules in this manner, we have
developed an algorithm that minimizes the fraction of genes or arrays
that appear as disconnected module fragments. Thus, our tool maintains
the aforementioned simplicity of the common cluster representation,
while allowing for an intuitive presentation of overlapping groups of
genes and arrays.
Transcription modules provide the building blocks of the tran-
scription network. A systems biology approach aims not only at iden-
tifying (and annotating) these “blocks”, but also describing the
relationships between them in order to reveal the structure of the
entire network. Module relationships can be defined in many ways: the
extent of common genes, conditions, or functional categories
describes static intermodule relations. Yet, our identification of mod-
ules in terms of gene scores and condition scores also retains infor-
mation on what experiments induce or suppress the module's genes,
and to what degree. Thus, correlating modules over these scores can
