Hello everyone
I am currently working on the genome annotation of a non-model green alga. For this project, we have sequenced the transcriptome using PacBio Iso-Seq (~10M reads, polyA-selected, no cap selection), and 12 paired-end short-read RNA-seq samples are available on NCBI (~20M reads per sample).
My plan is to use BRAKER4 for the initial annotation. There are a few existing annotations for closely related species on NCBI, but all of them were built from less data than we have, so I would prefer not to rely on homology-based methods.
This brings me to my question: given that BRAKER produces robust but still imperfect gene models, is there a way to refine the annotation using our actual transcriptome data?
After consulting several AI bots, the best suggestions I received were to use either TSEBRA or EVM to combine BRAKER output with collapsed Iso-Seq data and assembled short-read transcripts. Neither option seems right here: EVM is obsolete, and TSEBRA was specifically designed to merge multiple BRAKER runs rather than to integrate external transcript evidence.
I would be very grateful for any suggestions or clarifications on a better approach.
•
written
by
Andrei Makhon •
0 answers
No answers yet.
Log in to answer this question.
Best approach will be to dive in manually ;-)
use a genome visualisation tool in which you can edit models (IGV?, Genomeview, Apollo, ...)
any other approach will likely inject yet another level of uncertainty unfortunately
(or stick to Braker and try to improve the annotations by improving how you run Braker)
Best, yet the most labour-intensive :( Going through 15k genes and isoforms could take half of my PhD.
Any suggestions on how I can reduce the search space?
The only idea I have so far is to compare the BRAKER output against the collapsed Iso-Seq data and flag large structural discrepancies.
Not arguing on that point :) (on the other hand: you would be surprised how efficiently this can work out in the end ;) ) ... anyway
A very crude approach I often used to apply back in the days was to plot gene length ratios. Take the genes/proteins of a related organism (preferentially well annotated). Do an all-vs-all blast and plot the ratio of the best-hit vs query length. This way you can quickly get a feel of the overall picture and even zoom in on the genes that have severe gene length differences. Likely those of the ones with the largest mistakes in them. (You will miss the very small differences this way but they are usually of less impact). Just pay attention to not force all genes to have the same length as 'known' as protein lengths do differ.
Checking them against the iso-seq data is also a valid approach indeed, but with eg UTRs etc the picture might be less clear-cut (and the 'other genome' approach is an 'unrelated' dataset
Is there a close(ish) relative available in the genome databases? By now there should be something available that could be used for doing broad comparisons of gene models. If there is a RefSeq version available, use that preferentially.
The question "is there a way to refine the annotation using our actual transcriptome data?" sounds strange, because BRAKER4 does use your actual transcriptome data.