Contents

About the Gene Expression Atlas

The Gene Expression Atlas (GXA) is a semantically enriched database of publicly available gene expression data. The data is re-analysed in-house to detect genes showing interesting expression patterns under the conditions of the original experiment.

Users can search GXA to discover up- and down-regulated genes in numerous experimental conditions, from diabetes to spaceflight. Searching for a condition (e.g. breast cancer) will return genes displaying interesting expression patterns under that condition. Searching for a gene will show experimental conditions in which that gene is up- or down-regulated. Users can also search for a combination of genes and conditions, to find analyses in which the specified genes have been found to show interesting expression in the specified conditions. Further details on how to search and interpret GXA data can be found here and in this online training course.

Construction of the Gene Expression Atlas

GXA is based on data from a subset of gene expression experiments from the ArrayExpress archive. Production Team curators select eligible experiments and annotate samples to a high standard using the Experimental Factor Ontology (EFO). Gene-specific annotations are automatically obtained from the latest release of Ensembl.

Raw expression data from each Affymetrix microarray experiment is pre-processed in R using RMA (Robust Multichip Averaging) [1], via the oligo [2] package from Bioconductor. This generates a table of normalized expression values for each gene (or probe set), where each row corresponds to a gene and each column to an RNA sample. For experiments using microarrays from other manufacturers, this table is provided by the submitter of the experiment.

The limma [3] Bioconductor package is used to discover differentially expressed genes. For each experimental factor, we ask whether each gene's expression level in each condition is significantly different from the mean expression for all conditions. For example, in an experiment with samples from tumors and healthy tissue, we ask whether the expression level in the tumor samples is significantly up or down compared with the mean expression level for all samples. To read more about the statistical methodology employed by GXA, please see the GXA publication. [4]

References

  1. ↑ Irizarry et al. (2003) Biostatistics 4:249-264.
  2. ↑ Carvalho and Irizarry (2010) Bioinformatics 26:2363-2367.
  3. ↑ Smyth, G. K. (2005). In: 'Bioinformatics and Computational Biology Solutions using R and Bioconductor'. R. Gentleman et al. (eds), Springer, New York, pp. 397-420.
  4. ↑ Kapushesky et al. (2012) Nucleic Acids Research 40:D1077-D1081.

Atlas Resources