Unstranded new transcript from Cuffmerge leads to error while RSEM prepares the reference

Question

[Thibault D.]

Hi,

I am looking for a solution to the following situation :

If I use unstranded RNA-Seq data through Tophat, then Cufflinks, finally Cuffmerge, I have a new GTF file with a set of transcripts newly identified. As the input was unstranded, those new transcripts have no orientation in the merged GTF file (i.e. a point (".") on the 7th column instead of a plus ("+") or a minus ("-")).

However, the first step in the RSEM pipeline (labelled : "rsem-prepare-reference"), fails due to the point on the 7th column. It is waiting for either a plus or a minus, nothing else.

I use cufflinks/cuffmerge before RSEM, because RSEM does not detect new isoforms in RNA-Seq data.

Many thanks in advance for your advices,

Thibault

Answer 1

[Devon Ryan]

You can just replace the "." with a "+" or "-". With unstranded data it won't matter which you use, in fact. Something like awk 'BEGIN{FS="\t"}{if($7==".") {$7 = "+"} print $0}' foo.gtf > foo.fixed.gtf should work (I don't recall if you need OFS="\t").

[Thibault D.]

Thank you for your answer, Devon

I thought it was not right to add it that way, and that it was the reason RSEM did not add a random strand by itself. I'll consider this option next time.