Hi Folks
I would love to get some opinions regarding duplicate removal during variant calling for the AmpliSeq Comprehensive Cancer Panel.
I have noted some previous discussions on the topic on this forum, but I am experiencing some additional issues.
We have been doing tumor vs. normal samples, and do the realignment and base quality score recalibration steps (no duplicate removal) using GATK. We are currently testing a variety of somatic variant callers (Strelka, VarScan, MuTect, JointSNVMix, SomaticSniper).
Ion Torrent states the following:
"Marking duplicate reads is not appropriate for Ion AmpliSeq data, because many independent reads are expected to share the same 5' alignment position and 3' adapter flow as each other. Marking duplicates on an Ion AmpliSeq run risks inappropriately flagging many reads that are in fact independent of one another."
When doing visual inspection of (possibly) variant positions in IGV, a lot of the variants seem to have been called based on read depth that sure look like duplicates upon inspection. There also seems to be a trend for variant positions to be at the edges of these reads .
However, if I do duplicate removal, my average coverage drops from 300x to 9x.
The duplicate reads vs. true coverage depth obviously has quite serious implications for all variant calling statistics...
Any advice / opinions / debate would be much appreciated.
Best regards!
Fourie
I would love to get some opinions regarding duplicate removal during variant calling for the AmpliSeq Comprehensive Cancer Panel.
I have noted some previous discussions on the topic on this forum, but I am experiencing some additional issues.
We have been doing tumor vs. normal samples, and do the realignment and base quality score recalibration steps (no duplicate removal) using GATK. We are currently testing a variety of somatic variant callers (Strelka, VarScan, MuTect, JointSNVMix, SomaticSniper).
Ion Torrent states the following:
"Marking duplicate reads is not appropriate for Ion AmpliSeq data, because many independent reads are expected to share the same 5' alignment position and 3' adapter flow as each other. Marking duplicates on an Ion AmpliSeq run risks inappropriately flagging many reads that are in fact independent of one another."
When doing visual inspection of (possibly) variant positions in IGV, a lot of the variants seem to have been called based on read depth that sure look like duplicates upon inspection. There also seems to be a trend for variant positions to be at the edges of these reads .
However, if I do duplicate removal, my average coverage drops from 300x to 9x.
The duplicate reads vs. true coverage depth obviously has quite serious implications for all variant calling statistics...
Any advice / opinions / debate would be much appreciated.
Best regards!
Fourie
Comment