Biology Reference
In-Depth Information
like the plaid model
30
and gene shaving,
31
aim to identify only the most
dominant bicluster in the dataset, which is then masked in a subsequent
run to allow for the identification of new clusters. The SAMBA
(Statistical-Algorithmic Method for Bicluster Analysis) biclustering
method
32
combines graph theory with statistical data modeling. While
each method has its advantages and disadvantages,
33
a common drawback
is their scaling properties in terms of central processing unit (CPU) time
and memory usage when applied to large data.
1.4. Complexity of the Output and Visualization
Visualization methods like topographic maps
9
or hierarchical dendro-
grams have been useful for the interpretation of results from standard
clustering methods. As the size and complexity of the data increase,
meaningful organization and visualization of the extracted features
becomes an ever more important issue. Valuable information beyond the
simple enumeration of clusters can be gleaned from the data, including
higher-order relationships between clusters,
7,9,33,34
analyses carried out at
variable resolutions,
29,33-35
and the identification of hierarchical organiza-
tion.
33-36
Visual representations can help to elucidate relationships
between genes, conditions, or clusters,
7,9,33-35
and to integrate additional
information about functional annotations or regulators.
7,8,24,37,38
1.5. From Modules to Models
Transcription modules not only provide the basic building blocks that
characterize the structure of the genome-wide transcription program
under a variety of conditions, but they also supply a map for a more inter-
pretable characterization of transcriptional changes induced by novel
experiments. In particular, searching for coherent changes in expression
in a larger module, one may identify patterns that are too weak to discern
when considering each of its genes alone. For example, Mootha
et al
.
39
showed that the coordinate expression of a set of functionally related
genes was significantly altered in human diabetic muscle, even though
this effect was too subtle to be apparent at the single-gene level (see, how-
ever, comment in Damian and Gorfine
40
). Segal
et al
.
41
used expression
