[maker-devel] MAKER chooser algorithm

[Carson Holt]

Sorry for the slow reply. Your e-mail was marked as spam for some reason by the e-mail list software, and I only review the spam folder on the list server every few weeks.

For the first one, you must have at least one piece of bridging evidence (perhaps blastx) or the two groups of evidence are within pred_flank= distance of each other (default is 200bp), which would cause the groups to cluster and create one evidence block. For that reason AED ends up being better for the merged model.

For the second one, it is the Augustus ab initio model itself that bridges the evidence cluster and it receives a better AED than the split model. The evidence clustering algorithm used for scoring is simple overlap clustering (no edges get broken). But you can use deFusion (https://wjidea.github.io/defusion/ <https://wjidea.github.io/defusion/>) to post process MAKER results, it will do a more complex analysis to break weekly supported edges on clusters and rebuild models the where appropriate.

—Carson


> On Jul 20, 2018, at 6:19 AM, Timo Metz <timo.metz at googlemail.com> wrote:
> 
> Hey,
> 
> I am working on the improvement of an already existing annotation. I could find that sometimes MAKER would split or merge genes where it intuitively does not look correct when looking at the evidence. Please find two examples attached. The first track is the old annotation, the second track the new annotation, then there is RNA-seq data, proteins, repeats, snap prediction, augustus prediction. It is visible, that in both cases the evidence supports two genes, and one gene predictor in each case tends to create one gene where the other one creates two genes. I do not understand why in this case the gene is merged, if evidence and also one ab initio prediction support rather two genes. Are there any suggestions on how to solve this?
> 
> best
> Timo
> <Picture1.png><Picutre2.png>_______________________________________________
> 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/20180821/a5397ea4/attachment-0001.html>