GeneMANIA co-expression networks are derived automatically from GEO data series with GSE identifiers. For each data release, we download all GSE series with a minimum number of samples (at least 12 but more for some organisms) that come from a set of GEO platforms that we have pre-defined as measuring mRNA gene expression. For each GSE, we identify the corresponding PubMed ID, which we use to name the network and extract meta-data, and then we compute the Pearson correlation coefficient (r) between all pairs of genes. We then sparsify the network by setting to zero any r that doesn’t appear in the top 50 highest r-values for at least one of the pair of genes. This network then undergoes our normalization procedure described above.
