A FC-GSEA Approach to Identify Significant Gene-Sets Using Microarray Gene Expression Data - Advances in Computational Science and Engineering - page 116

expression profiles. Unlike other previous approaches where gene expression profiles
only are employed for identifying differentially regulated genes and then some addi-
tional steps are required for understanding their biological meanings, the GSEA ap-
proach combines the analysis step for finding differentially regulated genes and the
interpretation step for understanding their biological significance into an integrated
framework in a systematic manner. The usefulness of this approach in a variety of
contexts has been shown in several previous studies [2, 3]. In spite of their impressive
results, however, the gene ranking by the signal-to-noise ratio (SNR)[1] used for
original GSEA produces some limitations such that for a specific gene-set it selects as
significant ones either highly up-regulated genes only or highly down-regulated genes
only in the end, not both at the same time. Thus, this makes it hard to reflect the situa-
tions in which both highly up-regulated genes and highly down-regulated genes play
an important role, just like biological pathways.
To deal with this problem, in this paper, we investigate the method, FC-GSEA, to
employ Fisher's criterion (FC) [5, 6, 7] for gene ranking in the gene set enrichment
analysis. With the FC-based gene ranking, we make an attempt to discover signifi-
cant pathways which may not be found by original GSEA using the SNR-based gene
ranking.
This paper is organized as follows. Section 2 reviews and summarizes original
GSEA approach and its uses in various applications. Section 3 describes our proposed
approach, FC-GSEA, which employs the FC-based gene ranking method. Section 4
describes the experiments on Golub's leukemia dataset and also evaluates the results
in a comparative manner with original GSEA. Finally, Section 5 concludes the paper
with some discussions.
2 Gene Set Enrichment Analysis
As a computational technique to identify statistically significant gene-sets showing
differential expression between two groups, the GSEA approach has attained a lot of
attentions by many researchers, showing impressive results in a variety of appli-
cations [1, 2, 3]. Specifically, Subramanian et al. [1] demonstrated how the GSEA
could yield some deep insights into several cancer studies such as leukemia and lung
cancers. Also, E. Taskesen [2] used the GSEA for performing some cancer studies on
human mammary epithelial cell lines, and in [3], the authors employed the GSEA and
other computational methods to identify robust subtypes of diffuse large B-cell lym-
phoma and understand more effective treatment strategies.
A summary of original GSEA approach [1, 2] can be given as follows:
[STEP 1] Compute the enrichment score (ES) of gene-set
- Rearrange all the genes in the decreasing order of the signal-to-noise ratio be-
tween two groups
- Build candidate gene-sets by using a variety of biological resources
- For each gene-set, calculate Kolmogorov-Smirnov score on the ordered entire
gene list and take the absolute maximum ES.