Figure 2

Empirical p-value₁distributions of random genesets for breast cancer OS and lung cancer OS. Relative frequency density distributions are shown for breast cancer OS (A) and for lung OS (B). For each survival estimate (i.e. breast cancer OS, and lung cancer OS), the relative frequency density estimate with a bandwidth equal to 0.01 is plotted for each empirical distribution (i.e. Biocarta-like, GO-like, KEGG-like, and CSR-like). A uniformly distributed empirical distribution would result in p-values₁ at the same frequency across the entire range 0 < p < 1 (dashed line in (A) and (B)). Ordered plots of empirical p-value₁ distributions versus the uniform distribution are shown for breast cancer OS (C) and for lung cancer OS (D). The permuted p-values used to model each distribution (i.e. Biocarta-like, GO-like, KEGG-like, and CSR-like) are plotted against random p-values₁ from a uniform distribution. An x = y line would be expected if the empirical distributions were uniformly distributed.