MICROARRAY DATA ANALYSIS IN AFFYMETRIX GENE CHIP - Quantitative Medical Data Analysis Using Mathematical Tools and Statistical Techniques - page 149

3.3. Gene Selection
We used the probe rank approach to analyze the prolactin data. Since it
was an in vitro study with a small sample size, we use P a = 0 and P b = 100.
The probe level threshold was set as 0.05 to calculate the probe percentile
dierence and the gene level threshold was xed at 50% to determine dif-
ferential expression for each gene.
The approach identied 65 regulated genes in the 23k PRL versus the
control group. Among them, 55 genes were down-regulated and 10 genes
were up-regulated. Similarly, 63 regulated genes (28 down-regulated and 35
up-regulated) were identied in the 16k PRL versus the control group. Here
we use the prolactin gene identied from the analysis results to demonstrate
the advantage of the probe level data analysis over the gene level data anal-
ysis in Figure 5. The probe expressions of this gene showed most probes
with dierential expressions between the 23k PRL and the control groups.
There were 15 probes (probes 1-15) with percentile dierence of weighted
rank greater than 0.67. This observation indicates a main treatment ef-
fect (homogenous dierential probe expressions) occurred in the 23k PRL
group. In contrast, comparison of the 16k PRL arrays versus the control
arrays showed only 10 probes (probes 1-7, 10, 12, and 14) with dierential
expressions. Specically, probes 1-7 had a percentile dierence of weighted
rank greater than 0.9. The result suggests an interaction eect (expression
dierences depend on probes) occurred in the 16k PRL group. These obser-
vations are consistent with the gene structure. The 23k PRL is a wild-type
human prolactin in which its mRNA closely matches the probe set of the
human prolactin gene in the gene chip. As a result, the expression of the
prolactin gene was almost completely dierential in the 23k PRL arrays.
On the other hand, 16k PRL has a quarter of the PRL molecule truncated
(i.e., alternative splicing). Thus, in the gene array analysis, only partial
probes showed dierential expressions and this explains the occurrence of
interaction eect. Clearly, this example highlights the importance of probe
level data analysis. If the gene level data analysis is used, we may miss this
target gene. Even when we can identify this gene, it only indicates dieren-
tial gene expression status without knowing probe expression, which may
reveal useful biological information.
3.4. Gene Function Classication
We used the integrated bioinformatics tool to perform gene function
classification. The regulated genes were grouped according to similar