[maker-devel] Maker pass-through behavior

[Carson Holt]

> So my question is how Maker handles pass-throughs, exactly. Can it
> reject pass-throughs, or should it always use such models over any other
> data source? Is there any scenario were it wouldn't?

pred_gff is treated the same as any other ab initio prediction.  It is just
one among several candidate gene models.  The model that is kept is the one
with the lowest AED score (lower means better evidence match/support).  Any
model with no evidence support or AED=1 will be rejected (no evidence
support) unless keep_preds=1 is set.  There is also another score eAED which
takes into account protein reading frame (protein evidence must be in same
reading frame as the gene model).  If eAED =1 it will also cause models to
be rejected.



> I understand that Maker uses some internal scoring system to estimate
> the accuracy of an annotation - could that be a reason?

Possibly.  Look at the AED score of the pass-through model in the final
MAKER GFF3 to see what the AED score was.  If you want to send me GFF3 to
look at with a list of regions you are concerned about I can tell you more.
Also consider giving PASA results to est_gff as well to bias the scoring
algorithm to maintain those models (I.e. Model supports itself, which is
reasonable since these are EST derived anyways and not just ab initio
predictions).  Also the model_gff option will always keep an input model
(with or without evidence support) and will only replace it with something
else if that something else has a better AED score.
> 
> 
> 
> And a related question: Is there  a comprehensive documentation where I
> can get more information on the internal decision making process of
> Maker? Or do I have to dig into the code for that?

Look at these two papers -->

Holt, C., and Yandell, M. (2011). MAKER2: an annotation pipeline and
genome-database management tool for second-generation genome projects. BMC
Bioinformatics 12, 491.
Eilbeck, K., Moore, B., Holt, C., and Yandell, M. (2009). Quantitative
measures for the management and comparison of annotated genomes. BMC
Bioinformatics 10, 67.

Thanks,
Carson



On 9/13/13 4:15 AM, "Marc P. Hoeppner" <marc.hoeppner at imbim.uu.se> wrote:

> Dear list,
> 
> I have started using Maker to explore its use for a number of genome
> projects we are planning on running. One of the tools we intend on
> incorporating into our pipeline is PASA (Since we will be using Trinity
> etc). The (cleaned) output with predicted gene structures I would like
> to pass to Maker as pass-through annotation (I am optimistic that
> way...) - but I noticed that doing so does not always result in the
> incorporation of the PASA gene model into the final maker annotation
> track. Sometimes it seems to be superseded by an Augustus/Maker model,
> sometimes the region stays empty (even tho a protein alignment is present).
> 
> So my question is how Maker handles pass-throughs, exactly. Can it
> reject pass-throughs, or should it always use such models over any other
> data source? Is there any scenario were it wouldn't?
> 
> I understand that Maker uses some internal scoring system to estimate
> the accuracy of an annotation - could that be a reason? It would be a
> bit odd tho, since a lift-over from chicken (to our bird genome) seems
> to support gene models produced by PASA, yet they are nowhere to be
> found in the final models.
> 
> And a related question: Is there  a comprehensive documentation where I
> can get more information on the internal decision making process of
> Maker? Or do I have to dig into the code for that?
> 
> Cheers,
> 
> Marc
> 
> PS I have attached a screenshot of such an example - the green track is
> Maker with proteins + augustus (chicken models) + PASA pass-through of a
> cleaned-up gene structure file. (Orange: Cleaned ORFs directly from PASA
> output, Grey: PASA ORFs without cleaning, Dark red: Maker with proteins
> and trinity transcripts as EST evidence, Black: chicken lift-overs from
> EnsEMBL)
> _______________________________________________
> maker-devel mailing list
> 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/20130915/0e1bdbbd/attachment-0003.html>