What "high duplication rate" means

[Kleido]

Hi all,
Could someone define "high duplication rate" in ChIP seq data analysis?
Thks

[dpryan]

Am I correct in guessing that you ran FastQC on your reads and it showed a really high duplication rate? If so, I would typically consider that normal unless whatever you're pulling down is rather non-specific in where it's located/binds.

[Kleido]

Actually the bioinformatician who ran the analysis said that, so I'm just trying to understand since he couldn't explain it to me

[SNPsaurus]

That probably refers to PCR duplicates... that is, even though you may have 90 reads at a location, they are likely to be 90 copies of the same original DNA fragment and so should not be considered independent binding events of your protein to DNA. This happens when low-complexity libraries are heavily amplified, which is common for ChIP-Seq.

You can determine if duplicates are a problem because you would expect that reads start at multiple locations across a genomic region where your protein was cross-linked, because the DNA was sheared randomly. If the reads start at only a few locations and there are multiple reads at each of the starts, then those are going to be duplicates.

As dpryan mentioned, with ChIP-Seq perfectly good data can look highly duplicated, since there might be very high coverage of reads in a constrained space. So it takes some actual examination of how the coverage looks to tell if it is just stacked high or duplicated.

[dpryan]

Just to add to SNPsaurus' response, keep in mind that gauging duplication rate is difficult if you have single-end reads. Then, the maximum coverage of a single position after removing what appear to be PCR duplicates is twice whatever your read-length is. Of course, for Chip-seq, this is unrealistic, so unless you have paired-end reads you're probably better off ignoring PCR duplicates.

[Kleido]

Thank you dpryan and SNPsaurus, that's very helpful.