This gene network is based on a total of approximately 80,000 microarrays from Gene Expression Omnibus. We analysed 54,736 human, 17,081 mouse and 6,023 rat Affymetrix arrays. Principal component analysis was done on probe correlation matrices of each of four platforms (two human platforms, one mouse and one rat platform), resulting in 777, 377, 677 and 375 robust principal components, respectively. These components are called transcriptional components as they hypothetically describe regulatory mechanisms of the corresponding transcriptomes, such as transcription factors or hormones. The components jointly explain between 79 and 90 % of the variance in the data depending on the species or platform.
Many of these components are well conserved across species and enriched for known biological phenomena. We therefore combined the results into a multi-species gene network with 20,000 unique human genes and were able to utilize the components to accurately predict functions for the genes. Predictions were made against pathways and gene sets in different biological databases: Gene Ontology (biological process, molecular function, cellular localization), KEGG, Reactome, BioCarta. For a detailed description see Fehrmann et al. Nat Genet. 2015 Jan 12. doi:10.1038/ng.3173.
Functional genomic mRNA profiler
A computational method (Functional Genomic mRNA Profiling, FGM-profiling) that is able to ‘correct’ gene expression levels for major non-genetic factors. The residual gene expression signal (i.e. functional genomic mRNA profile) correlates strongly with somatic copy number alterations. For a detailed description see Fehrmann et al. Nat Genet. 2015 Jan 12. doi:10.1038/ng.3173.