Graphics Reference
In-Depth Information
feature embedded in the distance matrix. he right panel of Fig.
.
uses the whole
grey spectrum to represent the distance range of
-
only, which reveals the main
three-group structure. he use of nonlinear color mapping (for the distances), like
the one implemented in MANET (Unwin,
), can also resolve this problem.
An Example
15.5
Construction of an MV Display
Many of the microarrays in Dataset
have lots of missing values due to technical
issues and because different experiments studied different sets of genes in the yeast
genome. Two thousand genes with four hundred arrays that had relatively few miss-
ingvalues werethenselectedfromtheoriginal Dataset
,resulting inDataset
.Illus-
trated in Fig.
.
is the MV display of Dataset
. Pearson's correlation coe
cient is
used to measure associations between genes and between arrays, a common practice
in gene expression profile analysis. Average linkage clustering trees are then grown
on the two correlation matrices for genes and arrays. he relative positions of the
terminal nodes of the two dendrograms are then used to sort the corresponding cor-
relationmatrixmapsandthedatamatrixmap(thegeneexpressionprofile).hebasic
geneclusteringstructureandarray(experiments)groupingpatternscanbeidentified
using these tree-sorted matrix maps.
heenlarged, permuteddata matrix map usedforgeneexpression profiling is dis-
played in Fig.
.
. Red dots represent a relatively high expression of message RNA
for the gene-experiment combination, green dots indicate relatively low expression,
and black dots designate relatively little differential expression. Missing values are
coded in white, it is clear that many of the arrays (experiments) still contain some
missing observations. Such an MV display presents each gene expression profile as
a horizontal strip of color dots across all arrays (experiments), and the important
visual information is carried by the relative variations in color hues.
Withoutsuitablepermutationsthatsortrowsofsimilargenessothattheyareclose
together and place identical arrays next to each other, so that the relativity property
holds, an MV display is basically useless. Based on this two-way permuted display,
one looks for horizontal strips of genes that share similar expression profiles, and
vertical strips of arrays that exhibit close experimental results. he blocks of the two
directions illustrate the interaction patterns of gene clusters and experiment groups.
Allofthe numerical information isdisplayedinthis rawexpression profilemap(with
proximity maps for genes and arrays and corresponding dendrograms). Careful and
patient examination of these color maps can lead to valuable insights into the em-
bedded information structure.
Examination of an MV Display
As with other visualization tools, both proper training and experience are required
to extract as much information out of these complex matrix visualization displays as
possible. While examining complex MV displays such as those shown in Figs.
.
and
.
, several general steps should be taken:
