[maker-devel] augustus underpredicting

[Carson Holt]

I don’t think you can use the protein2genome option to estimate gene count. It will turn any alignment that matches at east 50% into a gene model. So you can get a lot of partial models which will inflate gene count. It’s good enough for training but not so much annotation.

—Carson



> On Sep 19, 2017, at 6:02 PM, Xabier Vázquez-Campos <xvazquezc at gmail.com> wrote:
> 
> Thanks Carson.
> 
> Last quick question. After the first run (before using the gene predictors) I ran fasta_merge to get an idea of the numbers I should be looking for.
> In summary, I got 14000 genes, only using Swissprot and a close highly curated reference genome to avoid any "fake" protein or partial proteins from draft annotations, plus assembled RNA-seq from my genome. 
> How should I consider this as a guide? (if I can do so) ... Is this a number I should be aiming as a minimum number of genes? maximum? something around that?
> 
> PS my genome (fungus) is 80+ Mbp and just 150 contigs so I expect very few possible fragments due assembly (seq errors aside)
> 
> On 20 September 2017 at 07:34, Carson Holt <carsonhh at gmail.com <mailto:carsonhh at gmail.com>> wrote:
> Gene predictors tend to over predict, so I would not take the high numbers given by SNAP and GeneMark as true counts. You will probably end up with something like 7-10k in the final results. But now Augustus is giving a higher count, you should be good to start running MAKER.
> 
> —Carson
> 
> 
> 
> 
>> On Sep 17, 2017, at 7:12 PM, Xabier Vázquez-Campos <xvazquezc at gmail.com <mailto:xvazquezc at gmail.com>> wrote:
>> 
>> I did it that way and AUGUSTUS is predicting a more reasonable number of genes, about 12500 in Maker, but about 19000 in the model assessment step.
>> In comparison, SNAP gives 16000 and GeneMark 19000.
>> 
>> I haven't found any reference about but, would it be a good idea to train Augustus over the masked genome instead?
>> Thanks,
>> 
>> 
>> 
>> On 12 September 2017 at 02:50, Carson Holt <carsonhh at gmail.com <mailto:carsonhh at gmail.com>> wrote:
>> BUSCO may be generating too few models. BUSCO also identifies classes of conserved short genes that may not represent enough training diversity for your organism. Try running MAKER in protein2genome or est2genome mode, and then train with those results.
>> 
>> —Carson
>> 
>> 
>>> On Sep 10, 2017, at 7:03 PM, Xabier Vázquez-Campos <xvazquezc at gmail.com <mailto:xvazquezc at gmail.com>> wrote:
>>> 
>>> Hi,
>>> I have been annotating a fungal genome as usual, using Busco-trained Augustus (in addition to GeneMark and SNAP), but for some reason, Augustus is predicting a mere 207 genes compared to 15-20k from the other two.
>>> I've never had this problem. The genome has an unusual repeat content close to 50%, not sure if that might suppose a problem.
>>> Has anybody come up with any similar issue?
>>> I also asked to Busco developers if they have any idea https://gitlab.com/ezlab/busco/issues/49 <https://gitlab.com/ezlab/busco/issues/49>
>>> Cheers,
>>> Xabi
>>> 
>>> -- 
>>> Xabier Vázquez-Campos, PhD
>>> Research Associate
>>> NSW Systems Biology Initiative
>>> School of Biotechnology and Biomolecular Sciences
>>> The University of New South Wales
>>> Sydney NSW 2052 AUSTRALIA
>> 
>> 
>> 
>> 
>> -- 
>> Xabier Vázquez-Campos, PhD
>> Research Associate
>> NSW Systems Biology Initiative
>> School of Biotechnology and Biomolecular Sciences
>> The University of New South Wales
>> Sydney NSW 2052 AUSTRALIA
> 
> 
> 
> 
> -- 
> Xabier Vázquez-Campos, PhD
> Research Associate
> NSW Systems Biology Initiative
> School of Biotechnology and Biomolecular Sciences
> The University of New South Wales
> Sydney NSW 2052 AUSTRALIA

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://yandell-lab.org/pipermail/maker-devel_yandell-lab.org/attachments/20170922/edabec82/attachment-0003.html>