threshold for duplicate removal?

[mard]

I was trying out Picard's MarkDuplicates to remove duplicate reads before SNP identification in our targeted resequencing studies but I discovered that Picard classes non-identical reads that map to the same genomic location (same start and stop) as duplicates and only one read is kept. So this means that only a read from one haplotype can be kept for each location.

We're thinking of testing if it might be better to apply a filter/threshold to keep a certain number of reads that map to the same location instead of discarding all except one. Just wondering if anyone has tried something like this?

[bioinfosm]

Thats an interesting point. I agree with you that only one haplo will be kept in such a filtering. I have been only filtering reads that map to multiple locations; but keep using the duplicates, and guess that brings the PCR-bias in SNP identification (Hets look 30-40% variant, not 50%)

[mard]

Thanks for the information. I'm pretty new to next-gen analysis so am wondering if it's recommended to remove reads that map to multiple locations before SNP calling?