Validation of MIMGO: a method to identify differentially expressed GO terms in a microarray dataset

Yoichi Yamada ,Masanobu Oshima,Ken-Ichi Hirotani,Kenji Satou,Haruo Sawada

doi:10.1186/1756-0500-5-680

Abstract

BackgroundWe previously proposed an algorithm for the identification of GO terms that commonly annotate genes whose expression is upregulated or downregulated in some microarray data compared with in other microarray data. We call these “differentially expressed GO terms” and have named the algorithm “matrix-assisted identification method of differentially expressed GO terms” (MIMGO). MIMGO can also identify microarray data in which genes annotated with a differentially expressed GO term are upregulated or downregulated. However, MIMGO has not yet been validated on a real microarray dataset using all available GO terms.FindingsWe combined Gene Set Enrichment Analysis (GSEA) with MIMGO to identify differentially expressed GO terms in a yeast cell cycle microarray dataset. GSEA followed by MIMGO (GSEA + MIMGO) correctly identified (p < 0.05) microarray data in which genes annotated to differentially expressed GO terms are upregulated. We found that GSEA + MIMGO was slightly less effective than, or comparable to, GSEA (Pearson), a method that uses Pearson’s correlation as a metric, at detecting true differentially expressed GO terms. However, unlike other methods including GSEA (Pearson), GSEA + MIMGO can comprehensively identify the microarray data in which genes annotated with a differentially expressed GO term are upregulated or downregulated.ConclusionsMIMGO is a reliable method to identify differentially expressed GO terms comprehensively.

Highlights

We previously proposed an algorithm for the identification of Gene Ontology (GO) terms that commonly annotate genes whose expression is upregulated or downregulated in some microarray data compared with in other microarray data
Gene Set Enrichment Analysis (GSEA) was conducted for each recovery time-pointderived microarray data pair (e.g., 0 min vs. 7 min, 0 min vs. 14 min, 105 min vs. 119 min, 112 min vs. 119 min) from the yeast microarray dataset synchronized by α-factor
When any row in the matrix showing an false discovery rate (FDR) lower than 5% in equation (2) was identified for each GO term, we determined that GSEA + MIMGO detected that GO term as a differentially expressed GO term

Summary

Background

Microarray technologies allow simultaneous monitoring of the expression of thousands of genes [1]. The other methods do not have such problems, these others require pre-specification (e.g., 1 for a disease and 0 for a normal) of microarray data in which genes annotated with a differentially expressed GO term are upregulated or downregulated, or cannot identify these microarray data [9,10,11,12,13,14,15]. GO terms that commonly annotate differentially expressed genes between each pair of microarray data are identified using a tool such as GSEA (Figure 2). We applied a simple fold change method to identify differentially expressed genes between each microarray data pair in a yeast cell cycle microarray dataset, and tested the statistical significance of GO term annotations to the differentially expressed genes [16]. We combined GSEA with MIMGO to identify differentially expressed GO terms in a yeast cell cycle microarray dataset

Methods

III IV V VI VII VIII

I II 1 III IV V1 VI VII VIII

Results and discussion