We have recently incorporated ‘gene attributes’ as a source of data to use when searching for related genes. These attributes include gene features like ‘the presence of a kinase domain in the protein sequence’ or ‘expression in brain’ and are a valuable source of functional data about genes (15). However, unlike interaction data, these attributes are defined for individual genes. Currently, the GeneMANIA algorithm represents each binary attribute as an interaction network in which all genes with that attribute are linked. In the results page, attributes are represented as a diamond node (see Figure 2 for an example). To accommodate the large increase in the number of networks considered due to this change, we added a small amount of L2-regularization to the linear regressions used by our network weighting algorithms.
