Biology Reference
In-Depth Information
Fig. 3.
The iterative signature algorithm (ISA) is an extension of the signature
algorithm and is designed to reveal hierarchies of coregulatory units of varying
expression coherence. This approach is applicable also in the absence of biologically
motivated seeds, in which case the iterative scheme is initialized by many sets of ran-
domly chosen input genes. The output genes determined by the signature algorithm
are reused as input. This procedure is iterated until input and output converge. Each
resulting transcription module is self-consistent: its genes are most coherently coex-
pressed over the module conditions, which in turn induce the most coherent expression
of the module genes. Modules at different resolutions can be obtained by changing
the coregulation threshold parameter.
same gene can function in several processes, which are induced under dif-
ferent experimental conditions. (2) Requiring only coherent gene expres-
sion over a subset of arrays allows for picking up subtle signals of
context-specific and combinatorial coregulation. Given the experimental
noise in microarray data, these signals may be too weak to be extracted
from the correlations over all samples that are used by many clustering
algorithms. (3) Since the ISA does not require the calculation of correla-
tion matrices, it is highly efficient computationally and is thus applicable
even to very large datasets.
2.2. Comparative Analysis
We used modular algorithms for a comparative analysis employing expres-
sion data from
S. cerevisiae
,
C. elegans
,
E. coli
,
A. thaliana
,
D. melanogaster
,
and
H. sapiens
, and found that functionally related genes are indeed fre-
quently coexpressed in these organisms.
35
We applied the signature algorithm
