Duplicate Reads in ChIP-Seq

[sboba001]

Hi all,

I recently did a ChIP seq assay and when I did the analysis I only lost 10% of the reads in my IP. Previously coworkers have done a similar ChIP-seq and they lost about 90% of the reads. My understanding is that duplicate reads from ChIP-seq tend to be PCR amplification bias and thus having low duplication levels is a good thing. The problem is that the ChIP with only 10% unique reads shows the expected change in binding in control vs treatment and the one with 90% of unique reads doesn't show any change. Has anyone encountered this before? Or have a any ideas?

[nucacidhunter]

My understanding is that duplicate reads from ChIP-seq tend to be PCR amplification bias and thus having low duplication levels is a good thing.

In ChIP-seq and some other types of library the input diversity is low. That means to obtain certain amount of output library for sequencing, PCR cycle is increased which results in high duplication level of existing fragments. I do not consider this bias. High duplicate in some instances, for example, shotgun sequencing for de novo assembly is undesirable. If genome is large enough, sequencing depth is appropriate for genome size and input is optimal, it will be an indication of poor library prep. High duplicate is normal in most ChIP-seq experiments because only a small region of genome is targeted for enrichment by pool down resulting in low diversity.

The problem is that the ChIP with only 10% unique reads shows the expected change in binding in control vs treatment and the one with 90% of unique reads doesn't show any change. Has anyone encountered this before? Or have a any ideas?

High percentage of unique reads in this case could be indication of poor and non-specific pool down. I wonder if you had any control pool down to check specificity of your IP?

[sboba001]

Hi nucacidhunter,

I always check the IgG when I do the ChIP, but that was not sequenced.

[fanli]

Non-specific pulldown wouldn't necessarily show up in your IgG control. Did you sequence an input lane?

[sboba001]

Non-specific pulldown wouldn't necessarily show up in your IgG control. Did you sequence an input lane?

Yes, my input files are clean compared to my treatment samples. The problem is that I don't see a change in binding in my treatment vs my control.

[fanli]

The pulldowns should be a subset of the input. Is it possible that the two were switched?

One of our collaborators had a similar issue a while back, and they ended up redoing the IP to get a much cleaner result. I'm not sure if they ever tracked down why it was so dirty the first time around though....

[biocomputer]

Similar to what's already been said, but low duplicate could be a good thing (lots of uniquely aligning reads associated with your pull down and peaks) or a bad thing (poor IP so your reads align everywhere). Do the peaks in your IP sample look big/robust/as expected? Is there much background in the non-peak areas? Also what was the overall number of reads in your current IP sample vs the one done previously by your coworker? If you have a lot more reads overall you could saturate your peaks and end up with more duplicates.