Chip-Seq Data

[suludana]

Hi,

I am starting to analyse my first Chip-Seq experiment (using an histone antibody).
I obtained thousands of peaks.
After using two filters (read number/peak and control sample no IP) I obtained approximatly 100 peaks per sample. Is it normal? Am I filtering too much?

Thanks

[kopi-o]

In case of histone modifications, regular peak callers may not be appropriate if the modification is diffuse and spread out over large genomic regions. Is that the case here? 100 peaks sounds like a small number. Perhaps your library has a relatively low enrichment over broad regions?

[Nix]

I'd recommend letting the data tell you how many peaks you have. Use an app that controls for overdispersion and calls p-values using say a negative binomial distribution and also control for multiple testing by converting the p-values to false discovery rates. By setting an FDR of say 1% (and possibly further filtering your list to those likely to have an effect say >2x) then you'll have something to work with that's based on measure of confidence given the noise in your data.

Check out DESeq and edgeR if you like working with R or USeq (which wraps DESeq) for something a little more user friendly.

Here's a tutorial I wrote http://useq.sourceforge.net/usage.html