which annotation I should for RNA-seq? Ensembl, UCSC or refseq?

Question

[bxia]

Does it really matter?

I took a look of the annotation file, they looked very different...

[GouthamAtla]

Which organism are you working with ?

If they looked very different, other than the naming conventions of genes/chromosomes, don't you think it matters ?

Answer 1

[Denise CS]

It's not surprising the annotation files are different. They are independent projects annotating the same assembly (e.g. GRCh38 in human, GRCm38 in mouse) but relying on different methodologies for calling genes and transcripts. In Ensembl, the annotation needs to be supported by biological evidence (mRNA, EST, protein, RNASeq reads). Ab initio predictions are not listed in the annotation file whereas you may have some predicted transcripts in the RefSeq set (those based on XM or XP entries). The Ensembl annotation is the Gencode annotation, a merge between automatically annotated genes with manually annotated genes by HAVANA. The latter collaborates with labs/other groups to experimentally validate some of their transcripts (see Howald et al) and to include pseudogene annotation from Yale (see these slides).

[igor]

An important clarification is that this mostly applies to human and mouse. The annotation sources may vary widely for other species.

Answer 2

[Steven Lakin]

Choose whichever is the most up to date for the organism that you're studying; for the popular model organisms, it won't matter, but for more rarely studied organisms, it will depend on which database is the most up to date for that model.

Answer 3

[Antonio R. Franco]

Yes, it matters, but only in a reduced set of genes that are being differentially annotated among the different platforms

It has been recently published a paper about it, but I cannot find it right now. Sorry

[Denise CS]

Would it be this one by Zhao & Zhang?

[Antonio R. Franco]

Yes, it is the same paper