PCR duplicates, Rna-Seq and GATK

[whataBamBam]

Hi all..

I gather from this thread on biostars.

http://www.biostars.org/p/55648/

Linking these threads on biostars and here..

http://www.biostars.org/p/14283/

http://www.biostars.org/p/47229/

http://seqanswers.com/forums/showthread.php?t=6854

..and from some other reading around the forums.. that there's a general consensus NOT to remove PCR duplicates when doing Rna-Seq (at least at the stage of differential expression analysis) - for the perfectly sensible reasons discussed in those threads.

Previously with Dna resequencing and SNP calling, PCR duplicate removal was part of my standard preprocessing pipeline. Fine this is different - this is Rna we are dealing with now.

Now I am interested in doing some SNP calling. I'm planning to use Samtools and also GATK in order to achieve a consensus set of SNPs by taking the intersection of the results from the two tools (since this has worked well for me in the past).

Now my question is this..

I know from past experience that GATK's UnifiedGenotyper won't actually allow you to run it on Bam files which have not had PCR duplicates marked. So I'm wondering what my best option is here?

What I'm thinking is to use the alignments without duplicate removal for differential expression analysis.. and then use a second set which have had the dupes taken out for calling SNPs.

Is that a good idea? does that make better sense for the SNP calling regardless of GATK's fussiness over input files? It does seem to make sense to me that we want identically mapped reads for the differential expression analysis but not for the variant caling. However I don't want to follow an illogical procedure / do bad science just because of GATKs requirements on input files i.e. I want my decision to be based on science not getting tools to run - I could always just use a different SNP caller for consensus (i.e. Freebayes).