Down-weighting overlapping genes improves gene set analysis

BMC Bioinformatics

Table 4 A sensitivity analysis using simulated data in the absence and presence of overlap between gene sets

Scenario	GSA	GSEA	PADOG Setup I	PADOG Setup II	PADOG Setup III
1	5e-04	0.0015	0.0121	0.0378	0.0067
2	0.0276	0.225	0.0113	0.0374	0.0059
3	0.0654	0.2539	0.0133	0.0397	0.0111
4	0.0103	0.1535	0.0018	0.0271	3e-04
5	0.0161	0.2352	0.0011	0.016	1e-04

The table shows the mean p-values for GS₁ (designed to be relevant to the phenotype) over 50 different trials in each of the 5 different scenarios. GSA and GSEA p-values do not change if genes in GS₁ are found in other gene sets as well. Results for PADOG are given in the absence of overlap (Setup I), presence of overlap between the genes designed to be DE in GS₁ and other gene sets (Setup II), and presence of overlap between the non-DE genes of GS₁ and other gene sets (Setup III). All methods used 1000 permutations to compute the two sided p-values for GS₁. Best values are shown in bold and second best are italicized.

ISSN: 1471-2105