What is the real meaning of relative enrichment/peak height of ChIP-seq tracks?

Question

[HyperEvo]

Hello everyone. I am analyzing an ChIP-seq peak set and I wonder what is the real meaning of the relative enrichment or height of called peaks. I think ChIP-seq is a method to detect protein enrichment sites in genome. My question is, why these enriched peaks have different height when we check it in bigwig track. In my view, peak height means the research get more protein bound fragments from cell population, and there are higher fraction of cell that have protein bind to a peak region. Higher peak means there are more cell that have target protein binding at peak site. So I defined the peak height as level of relative enrichment of target protein by my self. Here is a screenshot of encode genome browser, the read peak is taller than the yellow peak. I want to know whether it is meaningful to compare the peak height or relative enrichment in a same track?

Sorry about my poor english cause I'm not a native speaker. This is my first time to post on BioStar, wish to have reply from anyone, Thanks~

Answer 1

[ATpoint]

You typically never compare read counts between peaks but between samples. Peak height is a consequence of many factors beside protein abundance, such as GC bias, PCR efficiency / amplification bias and mappability of the region. These super high peaks are probably overlapping with known artifact regions. Here is a repository that curates blacklists, a link to the paper is in there as well. https://github.com/Boyle-Lab/Blacklist

I always remove peaks overlapping the blacklist for my species (given there is one for your organism). These regions are consistently (regardless of protein) attracting mapping artifacts etc.

[HyperEvo]

Thanks for you answer! Actually I have already filtered my peak according to blacklist by bedtools. You mean it's meaningless to compare one peak with each other in a same sample? I have saw a kind of k-means clustering method to classify peak simply according to signal density, can I apply k-means to my WT and Treated groups and compare both peak cluster change, as I didn't found differential binding sites using software. For example, A peak was classify to cluster1 in WT and cluster3 in Treated, can I conclude A peak become weak in Treated group? Thanks for your answer~

[ATpoint]

I personally think that comparing the same peak between conditions is meaningful and comparing peaks within the same sample is not.

[HyperEvo]

Ok, thank you so much~