[maker-devel] geneid (or alternative ab initio predictors)

[Fields, Christopher J]

That is nice to know; I’ll have to check the masking on this assembly to see if that is the problem (my guess is that it is).

Carson, re: geneid and ‘hints’, it looks as if geneid can take some hints such as BLAST HSPs (as well as other information), in the form of a GFF ‘homology’ file.  I assume it could take protein2genome/est2genome as well through the same route.

chris

On Mar 10, 2014, at 1:31 PM, Sajeet Haridas <sajeet at gmail.com<mailto:sajeet at gmail.com>> wrote:

One of the problems I have found with genemark is that it does not understand a soft-masked genome. Hence, the self training is incorrect. I have found marked improvement to genemark's prediction by running the training on a hard masked genome.


On Mon, Mar 10, 2014 at 10:05 AM, Carson Holt <carsonhh at gmail.com<mailto:carsonhh at gmail.com>> wrote:
Adding a new predictor can take some time.  It obviously requires some
coding.  It’s usually not too hard just to convert results to GFF3 and
then pass it in.  Integrated support is really only beneficial for
predictors that can take “hints” from evidence alignments (for example we
are working on EVM integration right now -
http://evidencemodeler.sourceforge.net<http://evidencemodeler.sourceforge.net/>).  If SNAP and GeneMark give
problems just drop them.  GeneMark really doesn’t work very good on
genomes with complex intron/exon structure (and I really wouldn’t use it
for anything but fungi).

Make sure you are also giving sufficient protein evidence.  Perhaps all
proteins from chicken and pigeon for example.  Then you shouldn’t find
loss of any true genes if just using Augustus.  Also try not to use gene
count as an indicator of performance.  The value is very deceptive,
especially if the genome assembly is fragmented.

Thanks,
Carson



On 3/10/14, 8:52 AM, "Fields, Christopher J" <cjfields at illinois.edu<mailto:cjfields at illinois.edu>> wrote:

>I have been running MAKER 2.31 using Augustus and SNAP on an avian
>genome.  Augustus gives pretty decent gene model predictions based on a
>custom model we have and the hints MAKER provides.  However, SNAP seems
>to throw out a ton of false positives; in many cases this appears to
>cause erroneous gene fusions.  Leaving out SNAP altogether however leads
>to a marked decrease in # models overall, which is worse.  GeneMark had a
>very similar problem (high # false positives) and thus no marked
>improvement, either when using with both Augustus and SNAP or with
>Augustus alone.
>
>I have been exploring using geneid
>(http://genome.crg.es/software/geneid/) as an alternative, based on some
>feedback on another project I worked with int he past.  This would be
>feed into MAKER using external GFF, but I wanted to see if anyone has
>tried geneid with MAKER first.
>
>Finally, how hard would it be to incorporate alternative callers into
>MAKER?  For instance, would it be possible to add these like a ‘plugin’?
>
>chris
>_______________________________________________
>maker-devel mailing list
>maker-devel at box290.bluehost.com<mailto:maker-devel at box290.bluehost.com>
>http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org



_______________________________________________
maker-devel mailing list
maker-devel at box290.bluehost.com<mailto:maker-devel at box290.bluehost.com>
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://yandell-lab.org/pipermail/maker-devel_yandell-lab.org/attachments/20140313/357a688a/attachment-0003.html>