High number of optical duplicates on MiSeq

[mareen_engel]

Thansk austinso,

I think denaturation at a higher temperature sounds like a very good idea to tackle denaturation problems. (we did the standard NaOH denaturation 5min at RT, then mix with cold HT1).
Just to be safe: for the TruSeq RNA libraries stored as dsDNA, you suggest to mix with NaOH, 5min RT, ice, mix with cold HT1, 2min 95deg, ice, pool, load.
Woudl you suggest doing the NaOH before storage in later library preps as they suggest to do with the SGP plate?

Thanks, Mareen

[austinso]

Thank you Phillip.
While I agree that we may sometimes tend to overrate the problem of duplicates in RNA-Seq, I think it's a slightly different thing here, where I do see 3 Mio duplicates exactly next to each other on the flowcell with overall duplicate numbers which in my eyes are not high enough to render this statistically possible? Usually we see some 0-low hundreds of those optical duplicates even in our low-complexity runs?
For me this sounds like a problem in library prep/denaturation/clustering, would you disagree?

Best, Mareen

We've seen similar things, and tended to ignore dedup'ing for our particular application, which drove everybody (well...just the keyboard warriors) apesh*t.

Doing library denaturation with NaOH for 2 min at 96C as recommended for TruSeq libraries (http://support.illumina.com/content/...15027983-c.pdf, p.36) did the trick in removing these duplicates.

A.

[mareen_engel]

Thank you Phillip.
While I agree that we may sometimes tend to overrate the problem of duplicates in RNA-Seq, I think it's a slightly different thing here, where I do see 3 Mio duplicates exactly next to each other on the flowcell with overall duplicate numbers which in my eyes are not high enough to render this statistically possible? Usually we see some 0-low hundreds of those optical duplicates even in our low-complexity runs?
For me this sounds like a problem in library prep/denaturation/clustering, would you disagree?

Best, Mareen

[pmiguel]

I don't know -- I don't really trust programs that claim to be able to find optical or PCR "duplicates". How do these programs know they are not just biological duplicates? RNA will have a highly non-normal distribution with the possibility of just a few transcripts making up a large percentage of the library. If there is any fragmentation/ligation bias in library construction then you will see lots of the same sequence popping up irrespective of any instrument/PCR duplication.

--
Phillip

[mareen_engel]

Dear all,

we've run a Illumina TruSeq mRNA nonstranded library (mouse brain lower end of input recommendation) on a MiSeq for quality control with v3 Reagent Kit, 2*75 PE. The run has very good QCs, good cluster density and some 22 Mio reads.

After pairing the reads, Picard MarkDuplicates reports more then 3 Mio optical duplicates (i.e. duplicates less then 100pixel apart on the flow cell, similar results for 10pixel) next to roughly the same amount of "real" PCR duplicates (50% of QC20 reads after throwing the optical duplicates).
If analyzing the SE data without pairing, the number of optical duplicates is reduced to 0, thus it seems not to be a cluster-read failure.

For me, this indicates that there's 3 Mio clusters on the flowcell next to a cluster with their reverse complement strand which seems to be much above chance level even with a high PCR-duplicate RNA-Seq?

Potential explanations put forward by the representative contacted are incomplete library denaturation prior to loading (thus partially dsDNA library molecules hybridize to the flowcell and build two reverse complement clusters) or low complexity of the library (however, 3 Mio sounds really above chance level for me, the comeplexity of the library doesn't seem to be that bad?)

Has anybody ever seen this? Any ideas what is causing this high number of optical duplicates?
Or is this number of duplicates simply expected on a standard RNA-Seq on low input (~200ng)?

Any ideas or recommendations are very much appreciated!
Many thanks!

Regards,
Mareen



Edit1:
Ps.: Not sure which heading this thread should go to, please move if you have a better idea. Thanks!
Edit2:
library PCR was 12 cycles