Full text of DOI:10.1093/bioinformatics/18.3.490

Vol. 18 no. 3 2002
Pages 490–491
BIOINFORMATICS APPLICATIONS NOTE
GeneANOVA—gene expression analysis of
variance
G. Didier, P. Br e´ zellec, E. Remy and A. H e´ naut
Laboratoire G e´ nome et Informatique, Tour Evry 2, 523 place des terrasses de
l’Agora, 91034 Evry, France
Received on July 13, 2001; revised on September 14, 2001; accepted on September 24, 2001
ABSTRACT
3 FUNCTIONALITIES
Summary: GeneANOVA is an ANOVA-based software
devoted to the analysis of gene expression data.
Availability: GeneANOVA is freely available on request for
non-commercial use.
In our software, the use of ANOVA tools requires first
the definition of the design (i.e. the set of factors). For
instance, for data from Sekowska et al. (2001), we have to
enter the four above factors of experiments, and the factor
Gene. The definition of the set of factors can be saved and
reloaded for future utilizations.
One of the modules of the software performs ANOVA
over the whole set of data and displays the classical table
(see window at the top of Figure 1). This can be done
for interactions of any order. In the table, the user can
select or unselect factors or interactions (the unselected
ones are not taken into account in the computation). This
last point is sometimes necessary for some designs, like
Latin Square for instance, where some factors and/or
interactions effects are confounded (Kerr et al., 2000). The
table can be saved in standard text format.
Contact: didier@genopole.cnrs.fr
1
INTRODUCTION
This Applications Note presents a software written in
JAVA, especially designed to explore gene expression
data. It includes an interactive bidimensional visualization
module and offers several classical features such as
transformations of data, classification facilities, standard
visualization (plotting gene against gene and array against
array) and ANOVA tools. Here, we focus our presentation
on this last point.
In analyzing gene expression experiments, the preceding
step shows how much a factor contributes to the variation,
but it does not allow to detect the genes that vary
significantly with a given factor. To achieve this task,
another module of the software performs ANOVA for each
set of measures corresponding to a gene and displays
the results in a new table. However the relevance of
the expression of a gene depends on two characteristics:
first, the part of its variation due to the considered
factor, and second, the significance of this part. We
use the P-value for this last characteristic. However,
since the software does not assess whether the modeling
assumptions underlying the ANOVA hold, the validity
of the precise P-value cannot be verified. Instead, the
P-value should be used as an approximate indicator of the
strength of the evidence for differential expression.
In order to visualize these two characteristics simulta-
neously, we have developed an interactive visualization
module. It displays a graphic in which each gene is
plotted as a point, the abscissa being the variation due
to the factor normalized by the total variation of the
gene, and the ordinate the logarithm of the P-value
(see window at the bottom of Figure 1). The inter-
face allows the user to easily locate what point(s) is
2
MOTIVATIONS—INTEREST OF ANOVA
Most microarray or SAGE experiments aim at comparing
variations of gene expression with biological conditions
or different cell types. However there are other sources
of variation in these experiments (spotting conditions,
arrays... ). One way of taking this fact into account is
to design more complex experiments than those with
only two measures per gene. For instance, Sekowska et
al. (2001), studied in Bacillus subtilis the variation of
expressions across two sulfur nutrients (methionine or
methylthioribose). The experiments were repeated on two
differents days, with two different amounts of RNA (1 or
1
0 µg) and one replicate for each spot. So, there are 16
measures for each of the 4107 genes, one per combination
of these four factors: sulfur, day, amount and spot.
ANOVA is a statistical tool (Fisher, 1954) well suited
to the study of such a problem. More precisely, it
allows an estimation of the contributions of each of
these factors, and possibly of the interactions between
factors, in the total variation of the whole set of measures.
Moreover, under statistical assumptions, see Kerr et al.
(
2000) or Fisher (1954), it gives the significance of these
contributions (P-values).
490
ꢀc Oxford University Press 2002

GeneANOVA
violet zone in Figure 1). Genes in the lists can be selected
by clicking on their names or by typing their prefix in the
text field above each list. The color of the points on the
graphic changes with the selection of the corresponding
genes. A menu enables the user to display the name of
the selected genes on the graphics or to attribute a new
color to the corresponding points. Clicking directly on a
point of the graphic with the right button shows a similar
contextual menu.
In the display zone, putting the pointer on a point will
display a tool tip containing the name of the gene and its
coordinates.
Double clicking on a point or on the name of a gene in
one of the lists opens a window containing a page resulting
from an HTML request on the considered gene. This
request can be set up to any HTML-interfaced database,
like SWISSPROT for instance, via Preferences menu.
The window of the visualization module contains a
toolbar which makes it possible to zoom and to save the
graphics in JPEG format.
The software is able to handle datasets of usual size in
microarrays experiments (several thousands of genes and
a number of conditions). By using it, we easily retrieved
the set of genes reported in Sekowska et al. (2001).
Fig. 1. Top: table displaying ANOVA results. Bottom: graphics
module window with relevant genes (i.e. genes with large parts of
variation due to the factor Sulfur and a high statistical significance)
located in the selected area.
REFERENCES
Fisher,R.A. (1954) Statistical Methods for Research Workers. Oliver
&
Boyd, London.
Kerr,M.K., Martin,M. and Churchill,G.A. (2000) Analysis of vari-
related to a given gene (or a set of genes) and vice
versa.
ance for gene expression microarray data. J. Comput. Biol., 7,
8
19–837.
On the right side of the image, two lists report the names
of the genes. The first one (list Tot.) contains all the genes,
and the second one (list Sel.) only the genes contained in
the rectangular subarea defined by mouse selection (the
Sekowska,A., Robin,S., Daudin,J.J., H e´ naut,A. and Danchin,A.
(2001) Extracting biological information from DNA arrays: an
unexpected link between arginine and methionine metabolism in
Bacillus subtilis. Genome Biol., 2 6, 1–19.
491