Picard MarkDuplicates - whole bam file marked as duplicates

[tdoniger]

Hi,

I ran an RNA-Seq data set using the RUM-pipeline to align the data. I tried to use the Picard's MarkDuplicates - and it tagged very single read in the bam files as a duplicate.
I used the following parameters for MarkDuplicates:
java -jar /private/software/packages/picard-tools-1.84/MarkDuplicates.jar I=RUM-sorted.bam O=RUM-sorted-dups_marked.bam METRICS_FILE=dups_metrics AS=true VALIDATION_STRINGENCY=SILENT

(I had sorted the bam using samtools- but it was not recognized by MarkDuplicates that is why I used AS=true)

In the dups_metrics file it lists the percent_duplication at 43%

Any ideas?

Thanks!
Tirza

[whataBamBam]

Hey tdoniger

Did you ever find the solution to your problem? I have the same problem and my metrics file only claims one percent!

I'm currently running 3 different solutions to the problem and i'm waiting for the batches to finish..

- use samtools rmdup instead
- add VALIDATION_STRINGENCY=LENIENT to my command string for MarkDuplicates
- don't bother to mark duplicates (I only did this becuase GATK requires it - my other pipeline will run happily without this step) and try to trick GATK into accepting my file by adding an @PG line into my header to say I ran MarkDuplicates (yeah I know this is probably not recommeneded I just thought I'd see what happened) i.e. use samtols reheader to take the header from the file that MarkDuplicates marked ever read as a duplicate in and put it onto the file I wuld of used as input.

My command was

java -Xmx2G -jar MarkDuplicates.jar INPUT=infile.sorted.bam OUTPUT=outfile.sorted.dedupe.bam METRICS_FILE=myMetricsFile

I'd also previoulsy sorted the bam using samtools

[tdoniger]

I thought because every line contained: PG:Z:MarkDuplicates - that these were the reads marked as duplicates. This is not the case. It is the flag set in the second column that indicates whether it is a duplicate or not.

try: samtools flagstat library_no_dups.bam
You can find the flags that represent the duplicates. See- http://picard.sourceforge.net/explain-flags.html

[whataBamBam]

I solved this and now I can't remember how. Really slack of me not to come back and post the solution but tommorow I'll check my pipelines and have a look

[tdoniger]

Hi,

Thanks! But I was trying to explain that I managed to solve it. I had thought that every line marked by "PG:Z:MarkDuplicates" was a duplicate, but really this was not the case. The duplicates are marked in the flag in the second column of the same file.

Best,
Tirza

[TonyBrooks]

Going off piste here, but 43% duplication is pretty high.
How much PCR did you do?

[tdoniger]

quite a bit. very little starting material