when should I merge in annotation transdecoder pipeline

Question

[Mike]

Hi all,

I'm not certain what's the best way to do this, so any help will be highly appreciated.

I have a merged gtf file that I created after doing rna-seq>STAR>stringtie pipeline with a refernce genome.

I also have other gtfs and files such as: RFAM DB results, exonerate (protein alignment) result, ab initio result from AUGUSTUS.

I then used the transdecoder tool to predict ORFs but I only used the stringtie merged gtf file. Is it better to try and merge all of my diferent inputs to get a larger and more descriptive gtf and then use transdecoder on it for the final results, or should I use transdecoder on the rnaseq pipeline resuls and then merge the result gff with the other gtfs I got from different type of evidences?

The goal is to create gene prediction models based on all this evidence and the input genome.

Thanks a lot.

[lieven.sterck]

Not sure if I can follow:

it seems you want to do gene prediction on a genome but if I remember correctly TransDecoder is used for ORF finding on transcripts.

[Mike]

You are correct, StringTie pipeline results in an assembled transcriptome on which I use transdecoder

I thought that finding ORFs within these transcripts can help me gather information regarding the genes but perhaps I am wrong.

I'm trying to create gene models with what I have, but I lack knowledge in this field.

If you have any recommendation on how to proceed with what I have (assembled transcriptome, exonerate protein2genome output, ab initio gene prediction, RFAM results and transdecoder) Id love to hear it, Thanks!

[lieven.sterck]

ok.

No, you're not wrong here, at worst a bit sub-optimal :)

Answer 1

[lieven.sterck]

When you have all those information already at hand I would propose to go for a 'real' gene prediction tool/pipeline. Moreover, given the data you already have it would make sense to use a tool that can integrate all those information levels.

Tools that come to mind are: MAKER, EuGene, EvidenceModeler, GLEAN, ... (and there are many others). Those should be able to integrate the info you have and making use of the genomic sequence they will result in a merged non-redundant set of genes .

[Mike]

Thanks a lot,

I tried using EVidenceModeler but encountered many problems and bugs so I started thinking about doing this manually. I'll look into the other tools, if you have helpful knowledge regarding them I'll be happy to hear it.

Answer 2

[BioinformaticBird]

Hi, maybe you should give MOSGA a try.