Read dupliation in RNA-Seq

[aquleaf]

Hi All.

I have a question about read duplication. Our studied tissue is kind of special. Most of mRNA encode a small number of proteins. So in the FASTQC result, we got a high duplication level (> 50%).

Should we remove these duplicated reads?

Wish your help! Thanks a lot.

Best

[chadn737]

What do you want to do with the data? If you are looking for differential expression, then I would advice against it. For highly expressed genes and shorter genes you will see more duplicate reads the deeper you sequence and this is not necessarily due to PCR duplication. This is real data then and not an artifact.

Now this situation is arguably different if you are trying to do a de novo assembly, in which case I could see a real benefit in reducing duplicates, although I do not know enough about de novo assembly to speak with any confidence on this.

[arkal]

If I may request, could you run a picard estimate library complexity on your bam and let me know the result? Because I'm apprehensive about the accuracy of the fastqc duplication results! My sample (100bp PE bam) gives me approx 70% duplication with fastqc but only 13% with picards ELC...

[kopi-o]

It depends a lot on whether you have single- or paired-end reads. For paired-end reads, obtaining fragments from the exact same positions for both read 1 and read 2 (=exactly the same insert size) is very unlikely to occur by chance. In that scenario I think one gains more by removing duplicates (to remove putative PCR artifacts) than one gains by keeping the small fraction of true reads.