Question: Peak Calling Biasing!!! [Macs14]
gravatar for Sukhdeep Singh
7.7 years ago by
Sukhdeep Singh10.0k
Sukhdeep Singh10.0k wrote:

Hi guys,

I got puzzled by the peaks called by MACS14 peak caller. Attached is the image where black bar on the top of peaks represent that the peak is significant (peaks file from Macs). I observe significant peaks in the 2nd and the 3rd track but not in the first one though the enrichment (height of the peak) is two-fold more as compared to other, 74 vs 46 and 37.

What could be the reasons, I ran peak caller twice to double check with different bandwidth (250,300), I get same list of peaks. Also, the control used for peak calling is an Mock-IP control with 10 Mil reads where as samples has around 20-22 Mil reads. Macs is doing the inbuilt read normalization (Small dataset will be scaled towards larger dataset.).

How can I confirm its true or not, if true then what are the possible causes.

Thanks for your input.

enter image description here

macs chip-seq • 5.0k views
ADD COMMENTlink modified 2.6 years ago by Biostar ♦♦ 20 • written 7.7 years ago by Sukhdeep Singh10.0k


I have also some strange results from macs. Here I only can see that there is different in the left region, where the firs track has more reads. I would remove these reads (left part) and run macs again to see whether you get that peak significant or not doing that. Maybe to remove 200 pb left flanked region is enough to test that.


ADD REPLYlink written 7.7 years ago by Lorena Pantano360

Hey Lorena, I got your point about the shoulder, though there is a large difference in the peak height as well. I am not sure what caused this, I will try this. If I normalize the read depth among two samples, then Macs performance improves only that I get little less peaks in total. Thanks

ADD REPLYlink written 7.7 years ago by Sukhdeep Singh10.0k

A peak's significance is estimated locally for each candidate peak according to the local values in the control region. You wouldn't happen to have a high signal there in Mock-IP control sample, would you?

ADD REPLYlink written 7.7 years ago by Leonor Palmeira3.7k

No, there is no high signal in Mock at that position, and if there would be then there should'nt be any significant peak in the track B and C, as the control is same for all of them. Concerning local noise of the sample itself, you can see the shoulder next to the peak in question for Track A and B looks more or less the same, still the peak calling difference is there.


ADD REPLYlink modified 7.7 years ago • written 7.7 years ago by Sukhdeep Singh10.0k

I think this behavior is caused by the local estimation of the significance as Leonor pointed out already. Probably, if you lower the cutoff for the peak significance you may see your missing peak. Alternatively, I can suggest you to try MACS2 with the --broad option which will join all those smaller peaks shown in the image.

ADD REPLYlink written 7.7 years ago by Fidel1.9k

Hi Fidel, I also think so but I am baffled because using same mfold and other parameters, lower enrichment gets marked at peaks but not the high one, so it points to the local estimation. Also, I will try MACS2, but in my observation Macs2 is very stringent and outputs few peaks. Thanks

ADD REPLYlink written 7.7 years ago by Sukhdeep Singh10.0k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1134 users visited in the last hour