Comprehensive Comparison of Gene Set Analysis Tools

Gene set analysis has enhanced the microarray data analysis field with biological insights. The first introduced and widely used Over-representation analysis (ORA) method, has the limitation of the requirement of a predetermined differentially expressed genes list. To overcome this limitation, distribution based analysis (DBA) methods were developed with different analysis steps and null hypothesis. To understand the advantages and limitations of these methods, we present a comprehensive survey and evaluate the performance for nine commonly used gene set analysis tools. Methods testing self-contained hypothesis generally have better sensitivity and specificity than methods testing competitive hypothesis. But most of the methods have bias towards larger gene sets with self-contained methods more severe. Therefore, better sensitivity and specificity is obtained at the tradeoff of bigger bias in self-contained methods, and vice versa in competitive methods. We propose a combined performance plot to compare these methods, among which GSA demonstrated superiority over others.