among markers [24] but may also be heavily influenced by a few highly-significant markers [54]. Finally, z-score methods infer enrichment based on deviation from a normal distribution that accounts for the size of each pathway [52, 55]; while these methods are sensitive and fast, their error rates have not been well characterized. Self-contained enrichment methods employ even more diverse statistical methods to combine the p-values within a pathway into an aggregated measure (Table 2). However, in the absence of large-scale power comparisons among related methods across several well-characterized data sets, the choice of a particular enrichment tool may be less important than understanding the relative strengths and limitations of these broader categories.
