GC content of duplicate reads only

[swNGS]

I have a duplicate level of approx 10% in a DNA resequencing experiment and my read depth drops off in a linear fashion towards approx 70% GC.
The assay is a SureSelect, done on an Illumina analyser.
I'm fairly confident that the poor coverage in highGC is due to some sort of PCR amplification bias, but was interested in looking specifically at the duplicate reads re their GC content.
How do I go about splitting the duplicates out of my bwa-aligned bam into their own bam so that I can look into this specifically?
Thanks, Chris

[volks]

if you use picard MarkDuplicates (REMOVE_DUPLICATES=FALSE) you can filter using samtools view -b -f 1024 to obtain only reads which are marked as duplicates in the flag field.

[swNGS]

I think at the moment my pipeline is converting sam to bam, then marking duplicates. Presumably I can split the dups out of a bam file rather than having to convert back to sam, splitting out the dups then converting to bam for all the other downstream analyses?
Chris

[jflowers]

volks suggestion works on bam (hence the -b option).

[volks]

yes, from bam to bam (-b).
there is hardly ever a reason to write a sam file to disk.