MarkDuplicate, coverage

mikyi · 08-24-2011, 03:23 PM

Hi,

I am using MarkDuplicate.jar of Picard tools to remove duplicate reads. Then I convert bam file to pileup file to extract coverage for each residue. In the middle of that, I have two kinds of bam, bam before MarkDuplicate.jar and bam after MarkDuplicate.jar. I generate pileup file for each bam file by using samtool's pileup. Then I get the coverage for each residue from pileup file. The coverage value is different between bam before MarkDuplicate.jar and bam after MarkDuplicate.jar? I want to use coverage value for CNV analysis, so it is very critical for my analysis. Could you let me know if someone has the same kind of work. Thank you.

Heisman · 08-25-2011, 04:45 AM

Maybe I'm misunderstanding, but if you remove duplicates you would expect to see less coverage. You should definitely use the coverage values after removing duplicates (your second bam file).

Calle · 11-20-2011, 11:32 PM

No question is asked here really but I gather you are interested in knowing whether you should keep working with the duplicated removed or not for your CNV analysis.

The answer to such a question would be dependent on, among other things, your depth of sequencing, your samples (single individuals or pools) and your sequencing strategy (i.e paired reads or fragments).

A really high depth (for example 10000x) would give loads of duplicate reads even if PCR-duplicates were not present, especially if you are sequencing fragments but also paired reads if mean depth is sufficiently high and your insert size distribution is not tight.

At insane mean depths (for example from enrichment studies) this will lead to all or close to all reads being treated as duplicates, which means you will only count each read once, which ultimately will mean that you have no chance of differentiating a CNV from normal sequence.

Thus, I would be careful about just assuming dip-removal to be good for all kinds of analyses.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Velvet Assembler: expected coverage versus estimated coverage versus effective covera	DMCH	Bioinformatics	1	11-30-2011 04:21 AM
Bug of Picard's Markduplicate	xiang	Bioinformatics	12	04-11-2011 09:35 AM
"nucleotide coverage" to genome feature coverage	sheremey	Bioinformatics	3	11-02-2010 11:24 AM
low 454 coverage combined with high solexa coverage	strob	Bioinformatics	7	10-07-2010 10:14 AM
Coverage	suludana	Illumina/Solexa	1	12-15-2008 10:51 AM

08-24-2011, 03:23 PM	#1
mikyi Junior Member Location: USA Join Date: Dec 2010 Posts: 2	MarkDuplicate, coverage Hi, I am using MarkDuplicate.jar of Picard tools to remove duplicate reads. Then I convert bam file to pileup file to extract coverage for each residue. In the middle of that, I have two kinds of bam, bam before MarkDuplicate.jar and bam after MarkDuplicate.jar. I generate pileup file for each bam file by using samtool's pileup. Then I get the coverage for each residue from pileup file. The coverage value is different between bam before MarkDuplicate.jar and bam after MarkDuplicate.jar? I want to use coverage value for CNV analysis, so it is very critical for my analysis. Could you let me know if someone has the same kind of work. Thank you.

08-25-2011, 04:45 AM	#2
Heisman Senior Member Location: St. Louis Join Date: Dec 2010 Posts: 393	Maybe I'm misunderstanding, but if you remove duplicates you would expect to see less coverage. You should definitely use the coverage values after removing duplicates (your second bam file).

11-20-2011, 11:32 PM	#3
Calle Junior Member Location: Sweden Join Date: Feb 2009 Posts: 5	No question is asked here really but I gather you are interested in knowing whether you should keep working with the duplicated removed or not for your CNV analysis. The answer to such a question would be dependent on, among other things, your depth of sequencing, your samples (single individuals or pools) and your sequencing strategy (i.e paired reads or fragments). A really high depth (for example 10000x) would give loads of duplicate reads even if PCR-duplicates were not present, especially if you are sequencing fragments but also paired reads if mean depth is sufficiently high and your insert size distribution is not tight. At insane mean depths (for example from enrichment studies) this will lead to all or close to all reads being treated as duplicates, which means you will only count each read once, which ultimately will mean that you have no chance of differentiating a CNV from normal sequence. Thus, I would be careful about just assuming dip-removal to be good for all kinds of analyses.