Graphics Reference
In-Depth Information
very similar groups of genes. he other genes and pathways have more complicated
interactions between activities. It is of course possible to further exclude pathways
and genes with simpler behavior, and to focus on the details of interactions of the
more active genes and pathways.
Other Modules and Extensions of MV
15.8
So far we have introduced the fundamental framework for matrix visualization in
the GAP (Chen, ) approach to the visualization of continuous and binary data
matrices, with corresponding derived proximity matrices. We have also presented
some generalizations, such as the su cient MV and the sediment, sectional, and re-
stricted displays. In practice,observed data can behighly complex, tothe degree that
basic matrix visualization procedures are not rich enough to comprehend the data
structure. In some situations, one may not be able to apply MV directly to the given
data or proximity matrices. his section discusses ongoing projects and future di-
rections that will make matrix visualization a more promising statistical graphical
environment. One important feature of the GAP (Chen, ) approach to matrix
visualization is that it usually contains four basic procedures: ( ) color projection of
the raw data matrix; ( ) computation of two proximity matrices for variables and
sample; ( ) color projection of the two proximity matrices, and; ( ) permutations of
variables and sample. Most extensions of MV are related to the first two procedures.
It is simple to adapt the aforementioned algorithms for the other two steps once the
first two procedures are fixed.
MV for Nominal Data
15.8.1
It is much more di cult to perform MV for nominal data than it is for binary data,
sinceonecanuseblack/whitetocode / ifthebinarydataisasymmetric,andtheJac-
cardandothercoe cientsforbinarydatainordertoderivetherelationshipsbetween
variables and between samples. here is no natural way to guide the color-coding
for multivariate nominal data in such a way that the color version of the relativity
of a statistical graph still holds (Chang et al., ). he derivation of meaningful
between-variable and between-sample proximity measures for nominal data is an-
other challenging issue. Chen ( )and Chang et al. ( )utilized the Homals (de
Leeuw, ) algorithm and developed a categorical version of matrix visualization
that naturally resolved the two critical problems.
MV for Covariate Adjustment
15.8.2
Quite oten, covariate data (such as gender and age) are collected in a study in addi-
tion to the variables of primary interest. When the effects of covariates are an issue,
covariate adjustment must be taken into consideration, much as in a formal statis-
tical modeling process. Wu and Chen ( ) introduced a unified regression model
Search WWH ::




Custom Search