Links between distributions, detection spectra and indicator functions

Copyright information:

Taken from "Threshold-free high-power methods for the ontological analysis of genome-wide gene-expression studies"

http://genomebiology.com/2007/8/5/R74

Genome Biology 2007;8(5):R74-R74.

Published online 8 May 2007

PMCID:PMC1929143.

To illustrate important connections between gene score distributions, detection spectra, and indicator functions, we selected four categories from the imatinib data. The category MAPKKK cascade' exhibits a heavy lower tail, exemplifying a low-proportion-high-effects enrichment of underexpressed genes. As expected, this category was detected by ZK and ZC. The category 'EGF receptor signaling pathway' has normal tails but is left-shifted midway between the center and the tails, consistent with an intermediate-proportion-moderate-effects enrichment of underexpressed genes. This category was detected by AD. Whereas these two examples receive indicator values near -1 because they are enriched in underexpressed genes, category 'hematopoiesis' exhibits a right-shifted distribution, implying indicator values near 1. Category 'development', identified at a slightly higher false-discovery rate, has a heavy lower tail as well as a right-shifted center, and exemplifies mixed enrichments (intermediate AD, CM, ZA, and ZC indicator values).