Different duplicate results produced by MarkDuplicates and EstimateLibraryComplexity

[novabc]

Hi:
I compared Picard's MarkDuplicates.jar and EstimateLibraryComplexity.jar on finding optical duplication rate using the same sam/bam files. MarkDuplicates always reported 0 as optical duplication rate, whereas EstimateLibaryComplexity.jar reported different numbers. I am wondering whether anyone has tried this. Which method is trustable?
Thanks.
Jason

[dakl]

Hi,

The algorithm for estimating the complexity is different in the two tools. We downsampled files that we had sequenced to saturation and none was trustwothy. One overshot a factor 10, one underestimated it a factor 10. Cant remember which was which now off the top of my head.

It's interesting though. It's solved mathematically exact, and empirically it works extremely poorly...

[novabc]

Thanks for the comment. I found out that the version of picard that I used (1.48) did not seem to calculate optical duplicates by MarkDuplicates (always 0), and in EstimateLibraryComplexity results, READ_PAIR_DUPLICATES always incorrectly had the same value as READ_PAIR_OPTICAL_DUPLICATES. I tried the newest version, 1.86. Now the results seem more believable. EstimateLibraryComplexity gave a higher number (~10% more) of duplicates. But the ratio of READ_PAIR_OPTICAL_DUPLICATES/READ_PAIR_DUPLICATES was quite close, in my sample, it was 58% by EstimateLibraryComplexity, and 56% by MarkDuplicates.
Jason