Fig. 6

Explanation and validation of the TAPAS algorithm filtering step. For a given query isoform (orange node) a set of other genes (blue genes) are identified having correlated expression levels in their isoforms. The GO terms of these genes are compared with one another (excluding the query isoforms gene) to obtain an average GOSS score (see Methods). In example a isoform-P belongs to a cluster with low GOSS similarity and this cluster is discarded. In example b isoform-Q belongs to a cluster with high average GOSS similarity. The cluster is treated as valid and can be used to help characterise the functional neighbourhood of the query isoform. In c the link width represents GOSS scores between genes, the red links are used in the TAPAS filtering step. We find in the validation d that the average GOSS score of a cluster to the query isoforms parent gene (blue links in C) is significantly higher for filtered clusters. Note the filtering was applied only using the similarities between the non-query members of the cluster (red links in 5C). The ‘random’ plot is a control where the clusters have been generated randomly to show a background expected GOSS similarity between a cluster and the query isoform