A Data Warehousing Approach for Genomics Data Meta-Analysis - Evolving Application Domains of Data Warehousing and Mining - page 143

Table 4. Comparison CDF versus RMA via elicitation of differentially expressed genes

Data Comparison CDF RMA Gene in common

GSE6475 with RMA normalization ACNE-NormalSkin 223 239 216

GSE6475 with Median normalization ACNE-NormalSkin 214 239 213

an integration of datasets in a meta-analysis would
be useful to confirm the relevance of individual
analyses on specific Nickel-allergy biomarkers or
on reconstituted skin reactions for instance. We
present a first test on pre-normalized expression
data and a second one on raw data as CEL files that
are normalized after combination. For differential
expression analysis, we used a one-way ANOVA
(Analysis Of VAriance) via linear regression
model, defined as in section “MICRO-ARRAYS
EXPERIMENTS”, to identify a set of differentially
expressed genes between two different groups of
samples, the treatment group (Test) and untreated
control (Control) group.
These tests show that combining data may in-
troduce much noise since in these meta-analyses,
we do not retrieve afterwards previous results of
individual analyses obtained from each dataset
separately. Furthermore results presented below
show that it is very important to combine raw CEL
data rather than pre-normalized data sets.

• control subject at 7h compared to 0h

• control subject at 48h compared to 7h

• patient subject at 7h compared to 0h

• patient subject at 48h compared to 7h

Here we compared meta-analysis results with
individual analysis by a one-way ANOVA. In
Table 5 we can observe:

• in the second column, the number of dif-

ferentially expressed genes elicited by in-
dividual analyses on GSE6281 data only in
each of the four cases,
in the third column, the number of dif- •

ferentially expressed genes elicited by a
meta-analysis combining pre-normalized
GSE6281 and GSE7216 data in each of the
four cases (we applied the RMA-based trans-
formation method presented previously),
in the fourth column, the number of dif- •

ferentially expressed genes common in the
two previous lists.

Integration of Normalized Datasets

We used the same p -value and fold change for
each of these comparisons.
We observe an important dissimilarity between
individual and meta analysis: a large number of

We had previously conducted individual analyses
on GSE6281 datasets in order to extract differen-
tially expressed genes on:

Table 5. Comparison of differentially expressed genes elicited by an individual analysis on GSE6281
and a meta-analysis on pre-normalized data of GSE6281-GSE7216 series

Comparison Individual analysis Meta-analysis Gene in common

control -7h/0h 219 353 142

control -48h/7h 15 59 7

patient -7h/0h 223 413 146

patient -48h/7h 3069 5238 2849