Can I use preproccesed bam files for Cufflinks?

[cbaudo]

This question has been asked here previously: http://seqanswers.com/forums/showthr...s+preprocessed

Similarly, I have trimmed, aligned and sorted using STAR, add read groups, and marked duplicates to my BAM file. I'm curious if this processed BAM file is suitable for input into Cufflinks or if I should use the initial aligned BAM file.

Secondly, if it is suitable, would it make sense to before local realignment around indels using GATK prior to Cufflinks?

Thank you for your time,
cb

[dpryan]

As long as you had STAR add the XS tags for cufflinks then the rest should be fine. I actually don't know whether cufflinks pays attention to the duplicate flag. I assume not, since marking duplicates in RNAseq data is typically not useful.

There's no reason to realign RNAseq reads before using cufflinks. You're not calling variants (and even there, local realignment is becoming questionable with more recent versions of GATK).

[cbaudo]

I've read conflicting information about marking duplicates, do you suggest that I don't perform this step? Note that I haven't removed the duplicated reads.

Also, when you say it's becoming questionable do you mean it doesn't produce a higher quality alignment or it can introduce errors?

Thanks for your advice.

[dpryan]

You absolutely should not mark duplicates. Any highly expressed gene/transcript will necessarily have many apparent duplicates that aren't actual PCR duplicates simply due to the gene/transcript being highly expressed.

I mean that it doesn't increase quality. If you go over to Brad Chapman's blog you'll find a very large number of comparisons of GATK/samtools/freeBayes/etc. with a variety of settings. It's increasingly the case the local realignment and also quality recalibration don't increase the quality of the output (they don't generally hurt it, but these steps can take a while).