Graphics Reference
In-Depth Information
b) Look for interactions between the array clusters at off-diagonal locations.
Various types of between-cluster correlation patterns with substructure are
also easily pinned down.
c) hearrays represent many different biological assays forvarious functions of
Saccharomyces cerevisiaeyeast,suchascell-cyclecontrol,stress(environmen-
talchanges,relevantdrug-affected),metabolic/geneticcontrol,transcription-
al control and DNA binding (http://transcript-ome.ens.fr/ymgv/). Different
biomedical assays activate and suppress expression patterns of certain func-
tional groups of genes. We need to integrate this biological/medical knowl-
edge with the numerical/graphical findings in
a and
b to validate known
information and moreimportantly to explore and interpret novel interesting
patterns.
d) Hierarchical clustering trees for arrays and genes also yield a partial visual
exploration of the data and proximity structure, but this is not as compre-
hensive as direct visualization of the two proximity matrix maps, since the
dendrograms only retain some of the information in the proximity matrices
from which they are constructed.
. For the row (gene) proximity matrix:
a) Similar procedurestothose described in
aand
b forarrays (columns) must
be repeated for the gene (row) proximity matrix. Of particular interest is the
dichotomouspattern ofthese
genes.heup-regulated(red)genesatthe
upperhalf andthe down-regulated (green) genesatthe bottomhalfofthe A
arrays are responsible for this dichotomous structure. We denote these two
clustersof genes as G
and G
here.Several small subclustersof genes within
G
and G
can also be identified along the main diagonal.
b) Itisnecessarytogoonestepfurtherandconsultvariousannotationdatabases
for more detailed interpretations of and explanations for the potential clus-
ters of genes identified this way. Some of the genes have not been annotated
yet. heir potential functions can be roughly determined through the pos-
itively correlated (up-regulated) gene clusters and the negatively correlated
(down-regulated) patterns.
. For the raw data (gene expression profile) matrix:
a) Many major and minor array groups and gene clusters were found in steps
and
. In step
, we use the raw data (gene expression profile) matrix map to
search for the interaction patterns of the gene clusters on the array groups.
It is also necessary to use vertical strips of expression profiles to contrast be-
tween array group structure variations and horizontal strips to distinguish
between gene clusterdistribution differences. With careful examination, one
can associate certain blocksofexpression intherawdata matrix with thefor-
mation of each arraygroup and geneclusterandthe between-group (cluster)
differentiation.
b) here are about
(
.
%) missing observations in this data matrix of
arrays with
genes. It is clear that the pattern of missing observa-
