Figure 1.

FAIME transformation of gene expression arrays into Gene Ontology space (Panel A) and Cancer Module Space (Panel B) improves overlap by three to four fold as compared to standard bioinformatics techniques as demonstrated by differential gene expression (Panel C) or Gene Ontology enrichment of the differentially expressed genes (Panel D). In this analysis, we compared 3 different prostate cancer datasets (Yu, Wallace, Taylor) for differences between tumor and normal prostate gene expression in 4 different analysis spaces: Gene Ontology (Panel A), Cancer Modules (Panel B), standard differential genes (Panel C), and Gene Ontology terms derived from standard differentially expressed genes (Panel D). Each Venn diagram displays how well how each independent mechanism set overlaps in each of these spaces. The bold percentage in each panel provides the percentage of overlapping terms as a percentage of the average mechanism set length. Panel C demonstrates conventionally generated differentially expressed genes using the Significance Analysis of Microrarrays with a FDR of 5%. In panels A and B we first transform the gene expression arrays into either Biological Process Gene Ontology space or Cancer Module space, respectively. Individual pathway terms were analyzed standard t-test and adjusted for multiplicity. Terms with a FDR ≤ 5% were retained. The Gene Ontology terms in Panel D were generated by enriching the differentially expressed genes in Panel C using the DAVID tool and retaining terms with a FDR ≤ 5%. Legend: *Only concordantly deregulated mechanisms across datasets are counted in FAIME (e.g. significantly up-regulated ones in cancer against each other, then significantly down-regulated ones, then union of the two groups).

Chen et al. BMC Medical Genomics 2013 6(Suppl 2):S4   doi:10.1186/1755-8794-6-S2-S4