MICROARRAY DATA ANALYSIS IN AFFYMETRIX GENE CHIP - Quantitative Medical Data Analysis Using Mathematical Tools and Statistical Techniques - page 139

pathways. The pathways can be ranked in accordance with the p-values.
Another rigorous approach for gene pathways is the chain reaction model.
The chain reaction model has been widely used in the engineering eld to
simulate chemical reactions that occur in combustion devices such as jet
and rocket engines 58;59 . Its application to gene pathways can be done by
treating the regulated genes as a set of reacting species and calculating the
species changes as the gene expression changes. Since the chain reaction
model uses a set of chemical reactions to describe gene-gene interaction in
gene pathways, it provides an alternative for pathway level analyses such
that parametric studies of various pathways and genes-gene interactions
can be performed in an eective manner. Overall, these tools depict biolog-
ical interaction among genes and provide insights to study associations of
the biological pathways with research outcomes (e.g., disease versus non-
disease or treatment versus control). In this chapter, we describe a tool for
presenting the results in a simple, eective, and self-explanatory format to
facilitate the transition from data analysis to biological interpretation.
2. Methods
2.1. Data Quality Assessment
This subsection describes a 2D image plot to examine array comparability
and to assist verication of dierentially expressed genes 60 . The 2D-image
plot eciently sums up the information instead of a scatterplot. Moreover,
the 2D image plot can be used as a supplementary tool for gene selection.
By using an invariant band as an exploratory criterion, the 2D image plot
can be used to help validate whether a gene is dierentially expressed.
2.1.1. 2D Image Plot
The 2D-image plot rst reduces the data dimensions by grouping data using
percentile cutos. The percentiles of intensity in each array are used to form
k groups with x% for the interval length. That is, let Q 0 , Q x ; : : : ; Q (k 1) x ,
and Q 100 denote the cutos for the k groups, where Q a represents the ath
intensity percentile and kx = 100. For every two arrays, their percentile
cutos are used to form kk groups. The relative frequency is then calcu-
lated to represent density in each subgroup (grid). Thus, the high volume
of data is reduced to kk. The 2D image plot is then applied to the
grouped data. The percentile cutos of the two array intensities are used
as the rst two-dimensions in the x-axis and y-axis, respectively. Density