[maker-devel] Use pass-through system to add missing genes

Fri Apr 27 05:57:01 MDT 2012

Hi Anastasia,

On Apr 27, 2012, at 2:43 AM, Anastasia Gioti wrote:

> Hi Carlson, 
> Thanks for your help!
> 
>> The way you proceed depends on why the genes are not there to begin with.  Are they not there because of a lack of evidence?  
> 
> It is a mixture of cases, and I can only look at some examples to say that. There are cases where all 3 used ab initio predictors provide models, there are blastx hits, or both blastx and protein2 genome, but no EST evidence, thus  no model is retained. i guess my default parameters could be responsible for these cases at least.
> 

This doesn't sound right.  If there are predicted models and blastx protein evidence overlapping them you should get a model retained.  I know for the EST evidence that it has to support a splice site before it will be promoted and I can't remember if protein evidence is the same but certainly if you pass back those protein2genome predictions and the original proteins as evidence then they will be retained as models.

>> If that's the case just adding the new fasta file should do the trick.  
> 
> which fasta do you refer to? The proteins file I use as evidence contains all proteins i can actually use.
> 

Yes using the protein fasta from the closely related species as evidence.  I think you said you've already done that right?

>> Or are they not there because an assembly error makes it impossible to get a logical model for the region (I.e reading frame breaks).  
> 
> This is not the case in general.
> 
>> Are there ab initio models already called in those regions that could just be promoted to the annotation tier?  You can test that one by blasting against the nonoverlaping_abinits.fasta files.
> 
> I have not done this, will do!
> 
>> 
>> For any of the cases described, you can provide the existing annotation set as the input in GFF3 format, and previous models will be maintained preferentially.  
> 
> You mean in a new maker run? is this possible with the old maker as well, not maker2, right?
> 

Yes, the original MAKER will do this.

>> If you know which ab initio predictions you want to add (I.e. the ab initio promoting scenario I descibed), you can provide those predictions to the use the pred_gff option and then set keep_preds=1 and they will be maintained even without evidence.  Attached is a script that would make selecting those easier.  It take the MAKER generated GFF3 and a list of predictions to keep (one name per line).  These might be the results of a BLAST analysis for example.  It will then return the GFF3 entries for just those models selected.
> 
> The thing is, for the few cases I have looked at, I cannot really decide which model is the best, and the 3 models from the ab initio predictors do not agree on the exact intron-exon junctions or the start and stop codons.
>> 
>> If the situation is more complex, just provide more detail, and I am sure we can help you come up with a plan.
>> 
> What i was thinking to do was to provide a gff file of alignments (eg by exonerate) to the proteins of the closely related species that i am  missing, and somehow keep the previous annotations and get the extra ones by this gff file. But how exactly maker should be run to do this I am not sure. if I want to keep the previous annotations I  need the gff file of the last maker run as input, but then how do I discriminate with the exonerate gff file? And which mode of rediction should be on, and with which parameters? You mention keep_preds=1  for the existing annotations, but how do i also promote evidence from alignments on the same way in the same run?
> Looks feasible though. Thanks again,
> Anastasia
> 

Let me just restate what you've said so that I can be sure that I am correct about what you've already done.  You have run Maker with SNAP, Genemark and Augustus using EST from a closely related species (passed to altest) and protein evidence from other fungi.  You are missing about 1,000 genes compared to the species that provided the EST alignments.  You say their is good evidence that these genes exist from the alignments and I assume by this that you mean the EST/protein alignments that Maker produced.

1) Is the closely related fungus annotated and if so have you included it's proteins in the evidence set that you provided to Maker.  If you haven't provided these proteins as evidence to maker then you should do this.  You can re-run maker passing your original models back through like this:

#-----Re-annotation Using MAKER Derived GFF3
genome_gff=original_maker_annotations.gff3
est_pass=1
altest_pass=1
protein_pass=1
rm_pass=1
model_pass=1
pred_pass=1
other_pass=1

#-----Protein Homology Evidence (for best results provide a file for at least one)
protein=proteins_from_closely_related.fasta
## OR it sounds like you've already aligned these with exonerate?
protein_gff=proteins_from_closely_related_already_aligned.gff

2) If you've already included those closely related species proteins but still didn't get the 1,000 genes, then take your nonoverlaping_abinits.fasta and blast them directly against your closely related proteins.  Presumably they don't hit too well because if they did they should have been promoted to predictions by Maker the first time, but here you can decide yourself what thresholds to allow to keep the abinit predictions that hit the closely related species proteins.  If you filter you blast hits the way you want and keep the names of the abinit predictions that pass your filter, then use the script Carson attached it it will generate a abinit precidtion GFF file with only the predictions you selected.  You can then pass those predictions back to Maker and force it to keep them and Maker will turn them from predictions (match/match_part) into gene models.

#-----Re-annotation Using MAKER Derived GFF3
genome_gff=original_maker_annotations.gff3
est_pass=1
altest_pass=1
protein_pass=1
rm_pass=1
model_pass=1
pred_pass=0
other_pass=1

#-----Gene Prediction
snaphmm=
gmhmm=
augustus_species=
fgenesh_par_file=
pred_gff=ab_init_predictions_rescued_by_blast.gff

keep_preds=1

Barry

>> Thanks,
>> Carson
>> 
>> From: Anastasia Gioti <anastasia.gioti at scilifelab.se>
>> Date: Wed, 25 Apr 2012 11:09:36 +0200
>> To: <maker-devel at yandell-lab.org>
>> Subject: [maker-devel] Use pass-through system to add missing genes
>> 
>> Hi, 
>> I  have a set of predicted proteins from the genome of a fungus annotated by MAKER  using EST data from a closely related species and 3 ab initio predictors  (snap iterativelly trained 3 times, genemark trained directly on the assembly and augustus with a model from a less closely related species), along with a set of fungal proteins. I am missing ~ 1000 proteins when I compare to the species i used EST data from, and there is good evidence from alignments that these genes exist. The question is how to proceed from Blast hits to actual gene models here. The idea would be to add these genes to the existing dataset, rather than reannotate the genome. I believe that reannotating it without any further evidence such as RNA-seq from the species itself would not change much,and i d rather stick with actual predictions that i trust and have used in subsequent analyses. The 1000 genes I can accept to annotate with a less stringent and reliable way than MAKER, I just want to add them so that the difference in gene count gets corrected.
>> I was reading the MAKER 2 paper and i was wondering if I can use the legacy annotations scheme to do it, by providing GFF3 of the alignments between the two species in the regions where genes were missed, but as i said, I would not like to reannotate the whole genome, and running MAKER2 might cause slight changes that i d like to avoid. Is this possible? First, is it possible to provide a Gff3 file of specific locations and not the entire genome alignment? (I guess so..) Second, how can I tag the existing annotations as 'not to be changed' or alternatively, tag the new models only? How should I run maker2, with which predictors on and which off?
>> Thanks, 
>> Anastasia
>> 
>> Anastasia Gioti
>> Post-doctoral Researcher
>> 
>> anastasia.gioti at scilifelab.se
>> anastasia.gioti at ebc.uu.se
>> 
>> http://www.ebc.uu.se/Research/IEG/evbiol/people/pages/Gioti_Anastasia/
>> 
>> 
>> 
>> _______________________________________________ maker-devel mailing list maker-devel at box290.bluehost.com http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org
>> <gff3_select>
> 
> Anastasia Gioti
> Post-doctoral Researcher
> 
> anastasia.gioti at scilifelab.se
> anastasia.gioti at ebc.uu.se
> 
> http://www.ebc.uu.se/Research/IEG/evbiol/people/pages/Gioti_Anastasia/
> 
> 
> 
> _______________________________________________
> maker-devel mailing list
> maker-devel at box290.bluehost.com
> http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org

Barry Moore
Research Scientist
Dept. of Human Genetics
University of Utah
Salt Lake City, UT 84112
--------------------------------------------
(801) 585-3543

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://yandell-lab.org/pipermail/maker-devel_yandell-lab.org/attachments/20120427/3bbaebd2/attachment-0003.html>