A Data Warehousing Approach for Genomics Data Meta-Analysis - Evolving Application Domains of Data Warehousing and Mining - page 133

• MGED Ontology (Whetzel et al., 2006) is Series GSE6281 and 7216 are described in

section “STANDARD META-ANALYSES OF
DNA MICRO-ARRAYS” below.
In this kind of data, it is quite typical to search
for up-regulated genes at 48h compared to 7h for
instance or up-regulated genes in IL20 exposure
compared to control on the basis of statistical
techniques. Such statistical results lead then the
scientist to numerous comparative investigations
not only on the same dataset but on similar gene
expression datasets produced by human skin
biological reaction experiments.
For instance, investigators in such a situation
would obviously try to ask questions like these
ones: '' Which are the genes that are differentially
expressed in Nickel allergy and are not expressed
in cytokine exposure? ” or '' If gene CCL19 is
over-expressed in Nickel exposure at 48h, is it
expressed in other Nickel allergy experiments
too? ''. Multiple information sources may be useful
for answering: datasets from the same laboratory,
public datasets as described above, scientific
publications on the same issues, etc.
After studying differential gene expressions,
clustering methods are frequently applied for
identifying similarities among genes or samples.
Gene clustering results still raise new questions
on cluster meaning. Semantic data like concepts
defined in domain ontologies like the Gene Ontol-
ogy (GO) are useful for annotating a cluster with
its most descriptive properties related to biological
processes or molecular functions. For instance,
a biologist may probably wonder for instance
“ Which biological processes can be most likely
associated with clusters including genes CCL19
and IL2RA? ”, or more widely “ What are the other
analyses in which genes CCL19 and IL27RA were
clustered together and what are the semantic an-
notations of the clusters? ”.
Answers to comparative questions as shown
above should be obviously more relevant if
queries took into account a wider catalogue of
experiments and connected sources of informa-
tion. While so-called meta-analyses techniques

a framework of micro-array concepts that
reflects the MIAME guidelines and MAGE
structure. It provides controlled terms to
describe a micro-array analysis,
OBI (Ontology for biomedical investi- •

gations) (Smith et al., 2007) models the
design of an investigation, protocols, in-
strumentation and material used and data
generated. The adoption of such standards
for the management and sharing of micro-
array data is essential and provides benefit
to the research community.

AMI overvIeW

Analyzing a unique experiment dataset like pre-
sented in the previous section is a first mandatory
step for evaluating a micro-array experiment.
But results are obviously more valuable when
compared to other experiments. This section is
organized in two subsections. In the first subsec-
tion, we first present some practical scenarios
that motivated the design of AMI. Then in the
second one, we describe our approach to provide
a semantic solution to comparative analyses.

Motivations

Statistical analyses on gene expression data are
mainly conducted in order to identify differentially
expressed gene between two conditions. Let us
consider the two following experiments:

1. the gene expression analysis on Nickel al-

lergy as described in (Pedersen et al., 2007)
and its datasets with kinetic gene expression
of Nickel exposure at 0h,7h, 48h and 96h
after exposure,

2. the gene expression analysis on psoriasis as

described in (Sa et al., 2007)and its datasets
with gene expression for exposure to 8 cy-
tokines such as IL20 or IL1b.