the Ensembl compara homology pipeline) to identify initial gene structures that appear to be evolutionarily inconsistent. These regions are then subject to a second, more computationally expensive localized gene build pipeline with more sensitive parameters. The major classes of problems identified are split genes, missing orthologous genes, partially predicted genes and false exons. For the test case of the horse genome with initially 20 322 gene models, this post-processing pipeline identified 236 genes that were split; added 1013 genes that had initially been missed, but for which there were orthologs; extended 1330 partially predicted genes and removed 840 false exons. The process is now being systematically applied to other high coverage mammalian genomes. These genesets will be patched in subsequent Ensembl releases.
