MICROARRAY DATA ANALYSIS IN AFFYMETRIX GENE CHIP - Quantitative Medical Data Analysis Using Mathematical Tools and Statistical Techniques - page 141

Fig 2.1 Fig 2.2

Fig. 2. Figure 2.1 is a 2D image plot for two replicate arrays, A1-A2. The large discrep-
ancy of intensity can be seen by a wide band in the diagonal curve. In contrast, Figure
2.2 shows high degree of consistency between another two replicates (B1-B2).

the normalized data, then it is possible that such dierence is caused by
over-correction in the normalization procedures. It is useful to verify a set
of selected genes in the raw data format (data before normalization) after
these genes are selected. The graphical approach described is useful and
ecient in displaying the relative dierence visually by comparing to the
invariant band. We can use the invariant band as an exploratory rule to
examine whether a gene has dierential expressions from one experiment
condition to another.
We illustrate using this approach as a supplementary tool for gene se-
lection by the example of a spike-in gene, CreX 5, in the four arrays E1-E4
(Figures 3.1-3.6). The 2D image plots using the 10% deviation invariant
band (i.e., m = 2; x% = 5%) are displayed to show distribution of probe
intensity for the spike-in gene. The pair-wise comparisons of the four ar-
rays show coverage rates of 83%-94%. The high coverage rates suggest these
arrays are comparable, and therefore the data are reliable for analysis. Ex-
amination of within group variations in Figures 3.1 and 3.6 shows that most
probe intensities of CreX 5 are located along the diagonal percentile curve.
This observation indicates that probe intensities of CreX 5 are comparable
between arrays E1 and E2 and between arrays E3 and E4 (i.e., small repli-
cate variation). On the other hand, examination of between group variation
in Figures 3.2-3.5 displays the majority of probe intensities far away from
the invariant band. That is, a large discrepancy of CreX 5 intensities occurs