at which step should I remove duplicate

[foxyg]

I tend to think I should run picard to remove duplicate after I use GATK to do realign and recalibrate, then remove duplicate. Or should I remove duplicate first.

I fail to understand the algorithm Picard uses to remove duplicates. Can someone explains how does Picard determine if a read is duplicate?

Thanks

[drio]

I don't think it should affect if you do it before or after (I do it before).

Picard looks for reads that are aligned in the same position and same direction and marks them as duplicates based on the basecall quality. If working with PE/MP data, both ends are taken into account.