A comparative study on gene-set analysis methods for assessing differential expression associated with the survival...

Area of Science:

• Genomics and Bioinformatics
• Cancer Research
• Statistical Genetics

Background:

• Existing gene-set analysis methods primarily focus on binary phenotypes.
• Performance comparison of Gene Set Enrichment Analysis (GSEA), Global Test (GT), Wald-type Test (WT), and Global Boost Test (GBST) for survival phenotypes is lacking.
• Two GSEA versions (GSEA1 with equal weights, GSEA2 with phenotype-correlated weights) were considered.

Purpose of the Study:

• To compare the performance of GSEA1, GSEA2, GT, WT, and GBST for survival phenotype analysis.
• To evaluate these methods using both simulation studies and real-world ovarian cancer datasets.

Main Methods:

• A simulation study was conducted varying gene correlation structures and the association parameter between survival and genes.
• Performance was assessed using power across different scenarios.
• Analysis was performed on two ovarian cancer datasets using a false discovery rate (FDR) threshold of q < 0.1.

Main Results:

• GT, WT, and GBST demonstrated consistently higher power than GSEA1 and GSEA2 in simulations.
• Ovarian cancer data analysis revealed GT, WT, and GBST identified 12, 6, and 8 significant pathways, respectively, while GSEA methods found none.
• Significant pathways identified by GT, WT, and GBST (Purine metabolism, Leukocyte transendothelial migration, Jak-STAT signaling) overlapped with prior microarray study findings.

Conclusions:

• GT, WT, and GBST exhibit superior power for survival gene-set analysis compared to GSEA1 and GSEA2.
• Gene correlation significantly enhances the power of these tests, particularly when survival is positively associated with genes.
• A synergistic effect for detecting significant gene sets exists when genes are correlated and survival association is unidirectional.