A Data Warehousing Approach for Genomics Data Meta-Analysis - Evolving Application Domains of Data Warehousing and Mining - page 151

GMineOnto To build GEOnto, we rely on:

• a corpora of experiment descriptions used GMineOnto provides concepts for the description

of models that result of data mining algorithms
application on gene expression data. In this first
stage, we focus on clustering models. Concepts
defined by GMineOnto describe the clustering
method, process and tool, and the set of clusters
discovered. External domain ontologies like GO
are linked with GMineOnto by the relationship
between a cluster and one of their most descrip-
tive concepts.As presented previously, the PMML
representation of a given cluster includes all se-
mantic annotations and their frequency into the
cluster. These annotations are thoroughly used for
cluster interpretation. GMineOnto relationship
between a cluster and an external concept allows
annotating a cluster with its most descriptive and
discriminative characteristics. Figure 10 gives a
fragment of the ontology where external concepts
are in grey.

to pick out candidate terms (e.g. “ bioSam-
ple ”, “ treatment ”, “ patch test ”, etc.),
dialogs with biologists for organizing the •

concepts and validate the proposed ontol-
ogy (e.g. “ patch test '' is a “ topical deliv-
ery ” which is a “ delivery method ” and a
“treatment” is “ delivered via ” a “ delivery
method ”),
existing ontologies UMLS and OntoDerm •

(Eapen, 2008) to extract specific concepts
(e.g. UMLS was used to enrich the concept
cell of the epidermis and OntoDerm to en-
rich the concept derm disease )

Figure 9 presents some concepts and relation-
ships of GEOnto.

Figure 9. Fragment of GEOnto