Does Picard tools Mark Duplicates consider read group?

[Heisman]

Hey all,

I ask as I'm trying to set up a fully automated pipeline where I anticipate having to merge data files in the future. If we sequence the same library more then once, we want to merge and then remove duplicates. If we sequence the same samples in different libraries, we want to remove duplicates and then merge.

But what if we sequence a sample once with one library, make a new library and remove duplicates before merging, and then sequence that second library again? Now, as the first two libraries are already mixed, it wouldn't be possible to just merge the second data set with the third and remove duplicates prior to merging with the first, unless all of the individual files are kept (of course if the sequence files are available it can be reanalyzed as such as well).

If Mark Duplicates handled different read groups (or libraries) separately within one bam file, this wouldn't be an issue as long as each library was given a different read group (or library tag).

So, that's where my question is coming from. Hopefully this situation won't arise but it'd be awesome if I could magically anticipate and build in code to handle every possible situation.

[Dzambi]

Any results

Hey Heisman,

Have you ever found out if MarkDuplicates was handling read groups properly?

[fchatonnet]

Hello,

having the same problem, I found your thread and I actually have the answer: MarkDuplicates handles read groups properly, based on the "LB" field of the "@RG" parameter, and not on the "ID" field, as one could imagine.
I just used it and it works perfectly, even outputing duplicates by read groups. So if you want to merge different bams and eliminate only duplicates of each bam (justified, since a read appearing in different runs is significant), be sure to indicate different "LB" fields in the "@RG" lines of your headers. Hope it'll help!