Statistical and Biological Evaluation of Different Gene Set Analysis Methods

AbstractGene-set analysis (GSA) methods have been widely used in microarray data analysis. Owing to the unusual characteristics of microarray data, such as multi-dimension, small sample size and complicated relationship between genes, no generally accepted methods have been used to detect differentially expressed gene sets (DEGs) up to now. Our group assessed the statistical performance of some commonly used methods through Monte Carlo simulation combined with the analysis of real-world microarray data sets. Not only did we discover a few novel features of GSA methods during experiences, but also we find that some GSA methods are effective only if genes were assumed to be independent. And we also detected that model-based methods (GlobalTest and PCOT2) performed well when analyzing our simulated data sets in which the inter-gene correlation structure was incorporated into each gene set separately for more reasonable. Through analysis of real-world microarray data, we found GlobalTest is more effective. Then we concluded that GlobalTest is a more effective gene set analysis method, and recommended using it with microarray data analysis