tag density v.s. binding affinity v.s. signal value

Question

[liu4gre]

I am wondering if it is OK to use signal value stored in narrowPeak format (ENCODE) to represent the binding affinity of a specific TF?

If yes, is this value comparable across samples?

Is it same to tag density?

If not, how can I calculate the binding affinity from narrowPeak file or bam file?

Thanks.

Answer 1

[Ying W]

signal.value from narrowPeak is from SPP peak caller

This value is not comparable across samples

It is related to tag density

You can count read overlapping the peaks, see this

Be sure to normalize after you count your reads (peaks have diff sizes and different samples will have different read depths)

See this for more information on what does signal.value mean for ENCODE narrowPeak files: https://sites.google.com/site/anshulkundaje/projects/idr

[liu4gre]

This is nice. Thanks. So it looks like only RPKM will be a good one to compare across samples...

[Ying W]

Not the best.. check out this paper

You might want to look into DiffBind or DBChIP if you are interested in differential binding

[liu4gre]

Great. Thanks a lot for the info.