I have a CLIP-Seq dataset I'm processing, which includes control samples and no inputs. This is the second CLIP analysis I've performed to help out users of our genome core facility and the first one only had inputs. I've done a number of ChIP analyses in the past, which I loosely consider to be a similar type of analysis, and they all used inputs, so I've never had to look into the differences in using inputs versus controls when doing peak calling before. Now that I'm trying to figure out what I need to do, I'm a little lost.

Here's what I've got:

  1. Samples of cross-linked RNA/protein complex that went through a poly-A and then a FLAG pull-down (protein G beads with an antibody). The complex was eluted, treated with proteinase K, and the RNA was sequenced.
  2. Controls went through the same processing as the samples except the control beads had no antibody

There are replicates in both cases. (As a side-note, the RNA was treated with RNase III while bound to the beads and the RNA's trimmed off the complexes were also sequenced. I'm not sure what to do with those samples yet.)

I was going to use one of the CLIP peak callers available on Galaxy, but whichever one I use, all the protocols I've found and the documentation I've looked at use inputs for the peak calling method and not controls. So what do I need to do differently (if anything) to use the controls as the background in the peak calling? Or do I need to suggest they also sequence the inputs?

And just so I know (dumb question here)... Is "input" (as the name suggests) the sample as it was before the FLAG pulldown? Why would one use one or the other (input versus control).

  • $\begingroup$ Did you get many reads for your control samples? I would think not, since there was nothing on the beads for them to even stick to. Normally this would be treated as input, i.e., not run through beads at all. What you have is roughly equivalent to using IgG as input in ChIP-seq, which generally works poorly due to few reads. What is the goal of the experiment? $\endgroup$ – Devon Ryan Aug 15 '19 at 7:00
  • $\begingroup$ There are roughly 10x fewer reads for the controls. I didn’t design the experiment. I’m just helping them out. I believe they want to know what RNA their protein is binding is binding to and where it’s binding (I.e. what motif). $\endgroup$ – hepcat72 Aug 15 '19 at 11:20
  • $\begingroup$ You'll probably just ignore the control samples then. $\endgroup$ – Devon Ryan Aug 15 '19 at 12:50

I've not handled CLIP-seq before, but we have done iCLIP and in this we had controls, but no input. @DevonRyan is correct, these control samples are similar to IgG pulldowns in ChIP seq.

The way we handled this was to call our peaks on isolated samples. The original peak calling alog from the first CLIP seq paper doesn't require input. Also I believe that clipper doesn't need input (although it will take it). We then removed any peaks from the sample that overlapped a peak in the control.

But most of our analyses we didn't do on peaks - we did lots of things like looking at metagenes, or classes of genes, or particular types of exon etc, and when we did this, because we were averaging over many sites, we could generate the pattern for the control as well as for the input, and often normalise the one to the other.

  • $\begingroup$ Is it similar to IgG pulldown though? The beads had no antibody, thus any reads coming out would be anything that just happened to stick to the raw (protein G) beads. An IgG pulldown would at least have something. As I understand it, an IgG antibody is non-specific? The controls have an order of magnitude fewer reads... So is there a way I can evaluate the effectiveness of the control? $\endgroup$ – hepcat72 Aug 16 '19 at 16:05
  • $\begingroup$ My understanding is that IgG is not non-specific, rather it is specific for the IgG of the host organism from which the primary antibody originated. Thus, in theory it shouldn't bind anything in the experiment, thus the IgG control in a ChIP pulls down two populations of fragments - those bound by a protein that interacts with igG (but shouldn't) and those that interact with the beads. Like CLIP, IgG controls in ChIP have much fewer reads. Despite this, you will almost certainly find that there are peaks in the control - there are definately sequences that interact with the beads. $\endgroup$ – Ian Sudbery Aug 19 '19 at 12:17
  • $\begingroup$ It seems straightforward that you could call peaks in the control and then exclude those peaks from your analysis of the actual sample - as they would represent likely background signal. $\endgroup$ – story Jan 13 '20 at 12:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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