Re: maker-devel post from guerrer@uni-duesseldorf.de requires approval

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Re: maker-devel post from guerrer@uni-duesseldorf.de requires approval

Carson Holt-2
Try looking at it in a browser.  You may see that models are merging because of evidence overlap, snap or augustus does not match the evidence well (you can drop one if it performs poorly).  But visualizing things will help.  For example fungi can lose genes because of UTR overlap (there is an allow_overlap option in MAKER for that).  mRNA-seq can falsely merge into extended transcripts (results in long UTR on models and maker has correct_est_fusion for that).  Neigboring parlors can cluster into a single model (you will see same hit coordinates multiple,e times in the browser - there is a tool called DeFusion for that).  Or evidence may be sparse (should be obvious from a browser).  Assembly errors can be found when you have nice transcripts alignments that do not have working ORFs (found by dragging and dropping est2genome results in browsers like Apollo and letting it calculate ORF).


—Carson


>
> Hello,
>
> It appears to me that my abInitio SNAP + Augustus (trained by nexflow) results are worse than the first (est2genome and prot2genome) run. The gene number decreases by more than 6000, the opposite of what I expected.
>
> Additionally, my abInitio rerun also has less genes than a previous SNAP reannotation.. The AED plots all look more or less the same but the number of genes lowers instead of increasing as I expected (image in annex).
>
> I'be been running my maker+nextflow pipeline with consistently good results, this species is the exception now.
>
> Do you have any suggestion? Should I ditch AbInitio training and use the Evidence based annotation? Or maybe just use the closest Augustus model instead of training my own?
>
> Kind regards,
> Ricardo
>
>
>
> <AEDs.png>
>

_______________________________________________
maker-devel mailing list
[hidden email]
http://yandell-lab.org/mailman/listinfo/maker-devel_yandell-lab.org