Software for prediction of long non-coding RNAs from RNA-seq data

[tigerxu]

I want to detect novel long non-coding RNAs from the mammalian RNA-seq data. Can you recommend some excellent tools for this analyses aim?

Thanks!

[syfo]

You could ask the people who did the GENCODE annotation, they must have a specific pipeline for lncRNAs. Make sure at least to take a careful look at the paper if not done already (Thomas Derrien et al, 2012).

Also, I'd check if the RNA-seq data experiment is compatible with this goal. For instance a high sequencing depth is needed because ncRNAs tend to have lower expression levels than mRNAs.

[tigerxu]

Thanks! Their paper didn't mention the tools/pipeline to identify long ncRNA from RNA-seq data. I will ask them if possible.

[blancha]

What's wrong with good old Cufflinks?
I know there are a lot of things wrong with Cufflinks.
But, what is so different about novel long ncRNAs, compared to other novel transcripts, that Cufflinks couldn't be used?

[syfo]

Right, Cufflinks does not require the data to come from coding mRNAs and will work as (good or bad as) usual. I think the specificity of the pipeline is after the RNA-seq data mapping/assembling. Once you have transcripts/transfrags/exons/whatever from Cufflinks or from another tool, how to identify novel ncRNAs?
Even if it does not overlap any known exon, a transfrag (or any set of overlapping RNA-seq reads) can come from an mRNA of a known coding gene (a new alternative exon, a part of an unusually long UTR) or from a new coding mRNA. Not mentioning that no guarantee can be made about the completeness of the transcripts you predict from RNA-seq data.
A last note: ncRNAs have lower expression values than mRNAs, so you might want to make sure the coverage of your RNA-seq data is compatible. Anyway I recommend a look at the methods and sup mat sections of the Derrien et al paper (Genome research 2012) and at the GENCODE standards for transcript annotation (manual curation, so high quality), although other refs might be more relevant.

[tigerxu]

Right, Cufflinks does not require the data to come from coding mRNAs and will work as (good or bad as) usual. I think the specificity of the pipeline is after the RNA-seq data mapping/assembling. Once you have transcripts/transfrags/exons/whatever from Cufflinks or from another tool, how to identify novel ncRNAs?
Even if it does not overlap any known exon, a transfrag (or any set of overlapping RNA-seq reads) can come from an mRNA of a known coding gene (a new alternative exon, a part of an unusually long UTR) or from a new coding mRNA. Not mentioning that no guarantee can be made about the completeness of the transcripts you predict from RNA-seq data.
A last note: ncRNAs have lower expression values than mRNAs, so you might want to make sure the coverage of your RNA-seq data is compatible. Anyway I recommend a look at the methods and sup mat sections of the Derrien et al paper (Genome research 2012) and at the GENCODE standards for transcript annotation (manual curation, so high quality), although other refs might be more relevant.

Thanks very much for your advices, syfo and blancha! I will notice these points.

[kongantik]

I wrote a pipeline for identifying long non-coding RNA if you have a list of known non-coding RNA. It worked well on my projects. If you don't have any, you can download them from various non-coding rna databases. Hopefully it is helpful to you.

http://git.io/SaFh1g

[tigerxu]

Many thanks for your help! I will use your script on my data.