After reading some of the forum posts in Biostar and SeqAnswers I find it very confusing whether to filter out the duplicate reads from aligned files or not. As far I understand it's very difficult to distinguish between highly expressed genes and duplicate reads and we may lose important information during the filtration process.

So, is it really necessary to remove the duplicates in differential expression analysis using RNA-seq data?

  • $\begingroup$ It all depends on how severe the problem is. In general you should only remove them if it is a very high percentage. $\endgroup$ Commented Aug 14, 2017 at 8:04
  • $\begingroup$ Any particular threshold like more than 35%? $\endgroup$
    – arup
    Commented Aug 16, 2017 at 17:25
  • $\begingroup$ Unfortunatly not. The problem is by removing them you might remove real duplicated data thereby under powering your analysis - but by not removing them you might count PCR artefacts thereby increasing false positive. So either option have a potential bias. I would probably try doing your analysis both with and without removal and then manually inspect some of the results only found by one approach to se which you find more trustworthy. What you should pay special attention to are whether the coverage is more uniform or seem to be affected by loads of reads from single positions. $\endgroup$ Commented Aug 17, 2017 at 8:42

3 Answers 3


For normal RNA-seq PCR duplicates are normally kept in, but the duplication rate can be used as a quality control: The higher the duplication rate, the lower the quality. For expression analysis, it is probably best to discard high duplication rate samples, rather than deduplicate them.

In general, the smaller the amount of RNA input into the library prepartion the worst the duplication. Many protocols for very low input quantities (such as single cell) include random barcodes called UMIs (Unique Molecular Identifiers). These allow PCR duplicates to be distinguished from genuinely independent molecules that just happen to have the sample mapping position.

  • $\begingroup$ Since the OP does not mention which RNA classes the analysis, let me add that UMIs are also extremely important for sRNA, in particular piRNAs, which quite often map to the exact same locations in the genome and have the same sequence. As afar I know current methods to mark duplicates in mapped reads would result in extreme under-reporting of these RNAs. $\endgroup$ Commented Aug 21, 2017 at 8:56

Generally you should just leave them as is. One does remove/mark duplicates in DNA seq.

For further read check this Nature paper

  • 3
    $\begingroup$ Your answer could be improved by hinting on why, according to the paper you cite, removing "PCR duplicates" shouldn't be done in RNASeq. E.g., by citing the one crucial sentence of the abstract: "We find that a large fraction of computationally identified read duplicates are not PCR duplicates and can be explained by sampling and fragmentation bias." $\endgroup$
    – BaCh
    Commented Aug 14, 2017 at 15:33
  • $\begingroup$ Sure..I could have excuse me for short answer. $\endgroup$
    – sbradbio
    Commented Aug 14, 2017 at 16:28

Some of the researchers in my institute have researched on this and use UMIs (Unique Molecular Identifiers) to eliminate PCR duplicates. They have been able to increase the reproducibility of the rna seq analysis they had been working on. The UMIs capture some lengths of diverse nucleotides from the rna-sequence. I cannot validate how much UMIs can help in deduplication, as even Nature journal says that UMIs specifically work on the 3’ end. But it has been observed that eliminating duplicates doesn’t improve the accuracy of quantification and it could introduce a bias in the data. So do not eliminate the duplicates, it's just a waste of time.

  • $\begingroup$ Hello Shirin, I think it's important to provide a reference for the following statement in your answer: "But it has been observed that eliminating duplicates doesn’t improve the accuracy of quantification and it could introduce a bias in the data." $\endgroup$ Commented May 29, 2019 at 13:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.