MICROARRAY DATA ANALYSIS IN AFFYMETRIX GENE CHIP - Quantitative Medical Data Analysis Using Mathematical Tools and Statistical Techniques - page 138

1.3. Gene Function Classication
In microarray studies, dierentially expressed genes are identied based
on descriptive statistics or a test statistics cut-o (e.g., p-value, false dis-
covery rate, or fold change) 36;37;38 . The selected genes indicate their sta-
tistical signicance of expression change, but researchers would also like
to know their biological relevance. Due to recent rapid developments of
genomic databases, the gene annotation information is now easily avail-
able from the internet. Various tools have been developed to incorporate
the information for gene expression proling. There are many tools avail-
able for searching and browsing Gene Ontology (GO) 39 , such as AmiGO 40 ,
EPGO 41 , GoFish 42 , Goblet 43 , and CGAP 44 . These are either web-based or
java-based tools to search for gene annotations. In addition, numerous tools
have been developed to map microarray data onto the GO structure, such as
GOstat 45 , eGOn 46 , DAVID 47 , GoMiner 48 , and FatiGO 49 . These tools can
be used to determine which GO categories are statistically signicant for a
list of genes, and to suggest the corresponding biological areas to warrant
further study. These tools also provide detailed analysis results, such as a
hypothesis testing for each GO category, clustering of functionally related
genes within a set of genes, interactive graphics, and numerous listings of
GO annotations for one or many groups of genes.
Moreover, investigators are interested in gene-gene interaction in par-
ticular pathways. Various pathway databases and methods (e.g., KEGG,
GENMAPP, REACTOME, CYTOSCAPE, and BIOCARTA) are available
on the internet for pathway analysis 50;51;52;53;54;55 . These curated databases
are useful resources to study biological processes, such as the pathways of
intermediary metabolism, regulatory pathways, and signal transduction.
They also help investigators gain insight into the potential functions of
new genes. Since the databases contain massive amounts of information,
it becomes challenging for researchers to convert the enormous amount of
information into useful knowledge. Many approaches have been developed
to provide parsimonious models to analyze gene pathways. For example,
MAPPFinder and Pathway-Miner are bioinformatics tools to create global
gene expression proles across biological pathways 56;57 . They classify genes
by integrating the gene ontology (GO) annotations based on metabolic,
cellular and regulatory pathways. Typically, a top list of genes selected
by one of these statistical methods is mapped onto pathways with gene
product association networks for genes that occur in the pathways. A z-
score or the Fisher exact test is then used to test statistical signicance of