Information Technology Reference
In-Depth Information
a background model:
X
1
L
S
i
=
[2 + log
2
(F
ij
)] ;
j
where F
ij
is the frequency of the i
th
base at the j
th
position. S
i
is in this case an
information-based measure of potential binding sites.
2.2.3. Integrated approach
A computationally ecient way to quantify the extent to which regulatory sequence
elements can explain changes in genome-wide expression data is to t the logarithm
of the expression ratio of a gene to the score of its promoter for a given motif's
PWM. The simplest case involve a single candidate motif which is tted on a single
expression condition e.g. using simple linear regression. If the motif has a statisti-
cally signicant regression coecient, than it is likely to be involved in determining
at least some of the observed expression changes. By increasing the number of
motifs and the number of expression conditions it would be possible to decipher
whole-genome regulatory networks. In this case, regression procedures involve a
great number of predictors (scores for all the motif considered) and response vari-
ables (all the expression conditions) and linear models are not appropriated. For
such a reason we might identify two procedures with good performance using high
dimensional data, such as Partial Least Squares [27] or Bayesian variable selection
methods [28].
2.3. Gene Coexpression Networks
Gene coexpression networks are intimately dependent on the regulatory network,
but they can reect subtle relationships that cannot be easily described starting
from the regulatory network. In Fig. 2.2 we show the simplest case of a direct
correspondence between regulatory and coexpression networks, common in Bacteria.
However, especially in eukaryotes, gene regulation can be combinatorial, so that
two genes controlled by a common TF can have several additional regulators, and
divergent expression patterns in dierent conditions, reducing the correspondence
between regulatory and coexpression networks.
2.3.1. Experimental approach
Microarray gene expression proling is a high throughput system providing large
amounts of quantitative data. The analysis of the data mainly looks at common
patterns within a dened group of biological samples, so that more focused ex-
periments are possible. Molecular data obtained can be absolute or relative gene
expression changes in a genome-wide fashion so that the transcriptome, the glob-
ality of mRNAs present in the cell at a given condition/moment, can be explored.
