Which set of genes does GREAT use?
Human and mouse
To limit the gene sets to only extremely high-confidence gene predictions, GREAT uses only the subset of the UCSC Known Genes<ref name="hsu">Hsu, F. et al. The UCSC Known Genes. Bioinformatics. 22(9):1036-1046 (2006). 1 Ashburner M. et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium, Nat Genet. 25(1):25-29 (2000). 2 . For zebrafish, we take the most upstream transcription start site.
|1||that are protein-coding (cdsStart != cdsEnd), are on non-random and non-haplotype chromosomes, and possess at least one meaningful Gene Ontology (GO) annotation|
GO includes information on the biological processes, cellular components, and molecular functions of genes. Thus, GREAT assumes that if a gene has been annotated for function at all then it is annotated in GO. Uninformative GO terms that do not allow entry into the gene set are 'Gene Ontology', 'biological process', 'cellular component', 'molecular function', 'obsolete biological process', 'obsolete cellular component', and 'obsolete molecular function'.
The zebrafish genome has no UCSC Known Genes set, therefore we used the following major transcript and protein source databases to obtain a comprehensive, high quality gene set:
We only included transcripts or proteins that belongs to ZFIN gene, because almost all ontologies map annotations to ZFIN genes.
As GREAT relies on high-quality mappings of genes to the genome to associate input regions with genes, we used stringent mapping parameters. All transcripts and proteins were mapped using BLAT requiring that at least 80% of the sequence matches with at least 95% identity to one co-linear locus in the zebrafish genome. These parameters are more stringent than the used in the mappings provided by the UCSC genome browser, which also annotates genes to loci where only a small fraction of the gene sequence matches. For GREAT, we need a higher stringency as inflating the number of loci for a gene compromises GREAT's statistical tests.
From all hits of transcripts/proteins in each of the three steps above, we retained only the best hit per locus, which effectively handles matches of paralogs. As a substantial number of bona-fide genes (such as Ctnnbl1 or Wnt9a) map to scaffolds, we include all gene-containing scaffolds in zebrafish GREAT. In contrast to the human and mouse gene sets, we also keep genes that currently do not possess a meaningful GO annotation because manual inspection found that the human ortholog often has annotations, indicating that zebrafish genes are simply less well annotated in GO. Furthermore, many of these genes have annotations in other ontologies. We expect that many of the genes currently without GO annotations will get annotations in the near future.
Our set of reliably mapped genes contains 14,214 genes mapped to 14,567 genomic loci for the danRer7/Zv9 assembly.
From the Saatvik: RefSeq transcripts that are associated to a ZFIN gene ID, our gene set contains 14,720 (95%).
The combined use of RefSeq and Ensembl transcripts substantially increased the number of genes that have annotations in our ontologies. If our gene set would be based only on RefSeq transcripts, we would miss 1,912 genes with annotations. Similarily, using only Ensembl transcripts, we would miss 1,218 genes with annotations.
How does GREAT determine a single transcription start site for each gene?
Many genes have multiple splice variants, however the vast majority of annotations available for these genes do not (and often cannot) distinguish between the different isoforms. Motivated by this observation, GREAT uses a single transcription start site to represent each gene in calculating gene regulatory domains. So, for human and mouse, GREAT uses the transcription start site of the canonical isoform of a gene. The definition of the canonical isoform is taken from the