Hi everyone,
Our team have processed DNase peaks data and discovered something strange that we're not sure how to interpret. The DNase peaks within same cell type containing some duplicated peaks with different signal values and p-values (please refer to the attached file). We're not sure how this event happened, and how to choose the proper value of these peaks. Are there some suggested ways to choose from these peaks? Thanks in advanced!
Created by Yi-An Tung yat I dont have the exact example you are referring to since the image you uploaded seems to be broken, but this is not unexpected from the peaks. Basically, in the TF peak calls (narrowPeak files) as well the DNase peak calls (narrowPeak files), you can often have peaks with the same chr/start/stop (column 1 3) but different scores (signal score, p-value) and different **summits** (last column). Basically, often you will find a contiguous region of enrichment defined by chr/start/stop with multiple local maxima (summits) within it representing nearby binding events. In such cases, the peak coordinates (chr/start/stop) will be the same but the summit locations will be different and their scores will be different. The summits represent the most likely location of binding events. So you can deal with this in different ways. E.g. center your prediction at the summit or take the maximal score for a peak across all its summits. The labels are binary so in the current challenge we just care about whether you predict a peak or not. We aren't evaluating the precise 1bp resolution predictions of binding events (which is harder to due because of lack of an objective ground truth of the true binding event at single nucleotide resolution).
If you paste your specific example here, I can confirm that what you are observing is multi-summit peaks.
Thanks,
Anshul. Your attached file/figure does not show up properly. Can you reshare it or email it to me at akundaje@stanford.edu
-Anshul