[maker-devel] annotation comparison

[Carson Holt]

Sorry for the slow response.  This was caught by my spam filter for some
reason, so I am only seeing it now.

Eval from WashU is the probably best tool for comparing two annotation
sets.  MAKER comes with a script that will convert MAKER's GFF3 output
into Eval GTF input.

If you are getting fewer genes than expected then you probably need to add
more evidence.  MAKER will be default reject genes that are not supported
by either a protein or an EST. The keep_preds=1 option makes it keep
everything even if there is no support.  Usually people make the mistake
of not providing sufficient protein evidence.  For example, just using
something like UniProt may not be sufficient, you may want to pick a
couple of other species with annotated genomes (fruit fly for example or
another mosquito species) and provide every protein from their genomes as
evidence.

Also if this is a newly sequenced genome, run CEGMA to see how complete
the assembly is.  If the assembly is 80% complete for example, you would
only expect to retrieve 80% of expected genes.

--Carson




On 8/8/13 9:11 PM, "Xiaofang Jiang" <jxf1023 at gmail.com> wrote:

>Dear Maker Developers,
>
>I am annotating a mosquitoes genome using Maker. I have two questions
>regarding Maker annotation.
>
>1. Are there any scripts available to compare two sets of annotations?
>We know about SOBA but were wondering if there is something more
>comprehensive that you guys use.
>
>2. I am expecting around 13,000 genes, however maker only predicted
>9,000 genes. I used both the gff3 from cufflinks, protein, and ESTs as
>evidence and SNAP as the ab inito predictor. I changed "keep_preds" to 1
>and that resulted 17,000 genes, it seems that shouldn't have happened.
>So in order to get more genes, should I try change "single_length" to
>100, and change "pred_flanks" to 100?
>
>
>Best,
>
>Xiaofang
>
>
>_______________________________________________
>maker-devel mailing list
>maker-devel at box290.bluehost.com
>http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org