Graphics Reference
In-Depth Information
tions is not a random one. Different array group and gene cluster combina-
tions have different proportions of missing observations. he visualization
of the missing structure is a great aid to users when they attempt to choose
amoreappropriate missing value estimate or imputation mechanismfor fur-
ther analyses.
c) Visual exploration provides valuable insights into more advanced studies,
such as the confirmation of existing metabolite pathways (see Sect.
for an
example) and the exploration of novel pathways.
his paragraph has only discussed some general issues associated with examining
an MV display. If we have access to expert knowledge and biologists familiar with
experiments related to Saccharomyces cerevisiae yeast, there are actually many more
interesting patterns that can be explored. In Figs.
.
and
.
we demonstrated an
MV can easily handle thousands of samples. An MV display can also handle thou-
sands of variables, since samples and variables are treated symmetrically in the MV
framework.
Comparison with Other Graphical
Techniques
15.6
In this section, we compare the visualization e
ciencies of the scatterplot (SP), the
parallel coordinates plot (PCP) (Inselberg,
; Wegman,
), and matrix visual-
ization (MV), based on varying dimensionality of the dataset.
Low-Dimensional Data
Forone-dimensionaldata,scatterplotandthePCPdisplaysamounttodotplots,while
a one-dimensional MV yields a colored bar chart. In any event, it is unlikely that any
method of displaying one-dimensional data will prove more popular than the his-
togram.Ascatterplotisthemoste
cientgraphicaldisplayfortwo-dimensionaldata.
While a PCP relies on the n connecting line segments between two vertical dotplots
to represent the association between the two variables, MV displays each sample as
a single rowwith two colored dots. he e
ciency of scatterplots decreases as dimen-
sionincreases.Forthree-dimensional data, arotational scatterplot iscommonlyused
toextractgeometric structurebyviewing asequence oftwo-dimensional scatterplots
over a range of angles controlled by the user.he usefulness of PCPand MV displays
of three-dimensional data is a subtle point, and the best permutation of variables is
definitely needed to enforce relativity for both types of displays.
High-Dimensional Data
A scatterplot matrix (SM) is used to simultaneously visualize the information struc-
ture embedded in all C
(
p,
)
pairs of variables for data with more than three di-
