A Data Warehousing Approach for Genomics Data Meta-Analysis - Evolving Application Domains of Data Warehousing and Mining - page 134

(see the next section) are able to involve multiple
datasets, they are essentially adapted to similar
micro-arrays datasets and do not provide semantic
enhancements.
The main goal in our approach is to provide
the investigator (most frequently biologist) with
enhanced semantic access capabilities to multiple
and huge sources of various data in order to lead
robust comparative analyses on gene expres-
sion data. We have thus designed a semantic
data warehouse which central component is a
knowledge base built on semantic annotations
and ontologies.
More precisely, we aim at providing func-
tionalities for multiple scenarios that may access
a variety of relevant sources of information. For
instance an AMI user should be able to:

expression are stored in a relational
database and clustering models are
described in XML files according the
PMML 12 formalism defined by the
Data Mining Group 13 (DMG).

◦ Annotation Storing Tools generate for

each relevant source of information,
specific semantic annotations based
on the domain ontologies. As illus-
trated in Figure 1 semantic annota-
tions are generated by different tools:
GEAnnot annotates experiments and
their experimental conditions (3),
GMineAnnot annotates synthetic
statistical data (5), MeatAnnot anno-
tates scientific publications (2) and
KnowAnnot annotates background

knowledge (1).

• select interesting publications, annotate ◦ AMI Querying Tools allow to navi-

gating into the knowledge base and
retrieve experiments, conditions or
genes according to more or less com-
plex criteria. This tool generates ex-
act answers and approximate answers
extracted according similarity links
in ontologies or deduced answers ob-
tained by logic inference rules.

them and store such annotations,
store explicit background knowledge and •

annotate it as facts and rules,
enhance previous statistical and data min- •

ing analyses on downloaded public raw
expression datasets and store results as
annotations,
store all expression datasets on a new local •

experiment, lead statistical and data min-
ing analyses on them and store results as
annotations,
lead meta-analyses on comparable selected

framework

• In AMI, thanks to underlying ontologies, each

available source of relevant information on a
genomic experiment is represented as a set of
semantic annotations. A search engine relying on
semantic ontological links ensures powerful and
intelligent querying functions which may retrieve
interesting and approximate answers to a query as
well as inferred knowledge deduced from logical
rule annotations.
Ontologies are formal representations of a set of
concepts within a given domain and relationships
between those concepts. They allow to reason
and to draw inferences about the properties of
that domain. In AMI, we re-used some existing

datasets and store results as annotations,
query the • AMI data warehouse for compar-

ative analysis by confrontation of knowl-
edge from all related sources.
◦ AMI Analysis Tools allow the us-

ers to process data transformations
for further combined meta-analysis
and to run statistical and data mining
tasks such as differentially expressed
gene analysis or co-expressed genes
clustering on relevant subspaces of
the data set. Such results on absolute
gene expression and differential gene