Picard MarkDuplicates per chromosome

[smleighton]

Is it ok to split an aligned bam by chromosome and then run MarkDuplicates on each of the files?

[Heisman]

No in the sense you'll miss PE reads that have each read map to different chromosomes. This is a key advantage of using MarkDuplicates over the samtools method.

[bruce01]

I wouldn't worry about splitting by chromosome and running MarkDuplicates, in fact seems quite astute to me. Do you expect gene duplicates and this is why you are doing it? Or is it due to limited computational resources, or another issue?

I have not tried this but would be interested to see if differences occur between split-by-chr and full bams, can you report back about it if you do that?

[bruce01]

No in the sense you'll miss PE reads that have each read map to different chromosomes. This is a key advantage of using MarkDuplicates over the samtools method.

What do these reads tell us? That there is a gene duplication? Would you retain these? And are they common? I have not found any in my own data, and would certainly remove them. I have only dealt with 100bp PE data though.

[smleighton]

Ive been asked to make a pipeline run faster on a distributed system. We can parallize most of the steps (alignment, some of the gatk steps, etc) but the MarkDuplicates is quite time consuming. If I split by chormosome then run on multiple machines it gets done much faster but I don't want it to affect the results.

I am planning tests to compare the deduped bams that are produced soon. I will post the results.

[Heisman]

Read through this FAQ for possible tips/explanations to make it faster: http://sourceforge.net/apps/mediawik...=Main_Page#FAQ

bruce01, no idea how common they are but if you want to remove the duplicates you can't split the bam file up by chromosome, I don't think (I could in theory be wrong; I've never considered doing this).

[volks]

I don't want it to affect the results.

you will not have the exact same results. the likelihood that will affect your conclusion is pretty low, it also depends on the question.