histone chip-seq profiles look similar to input! - (Feb/25/2013 )
Hi everyone, I'm having a problem here - I recently performed a chip-seq on very small amounts of mouse tissue for histone marks : H3K27ac and H3K4me1. Both of these showed very good enrichment on qPCR ( between 10-15 fold enrichment) when normalized to input (%input) and to negative sites. The library also looked very good when analyzed on the bioanalyzer. However, the enrichment profile (post-sequencing) for both markers disappointingly look similar to the input profile. I can see enrichments in many places (i.e. the positive control region) but they are not 'high' enough from the background levels according to my core here. Some of these are not seen in the input profile as well. To me, they might represent bona fide enrichments masked by high background levels.
Have anyone experienced similar problems before? Will appreciate any help/advice.
Did you include igG controls? Probably igG contorls are what you should compare to. But not all ChIP-seq experiments include igG controls. Input usually give you uniformly signal along chromosomes and you use input to normalize your signal against genome background. Yes, certain locations have higher signal in input than other places depending on sequence features and chromosomal structure. Have you determined the FDR cutoff based on your validation results? If not, you should determine the FDR and only consider those peaks that pass the cutoff as real peaks for further analysis.
Thanks pcrman - really appreciate your advice. I included IgG only for the qPCR analyses but not for the actual sequencing itself since the chip samples were highly enriched over IgG (in all forms of analyses). And we were told by the core that including the IgG will not make a difference at all. But please let me know if you strongly think otherwise. I might just include an IgG in my sequencing round if it's compelling enough to help reduce background noise (?). I will check the FDR that they've used and ask them to increase the stringency as you suggested. What is the normal FDR cut-off btw? Feel free to advise on any other bioinformatic technique that I could use to yield good signal/noise.
For each peak called by peak calling algorithm, there is a FDR assigned, you can randomly select a few peaks (ig, 10) from each of several ranges of FDRs for PCR validation. Based on the validation result, you set a FDR cutoff. You don't need to redo igG chip-seq, but igG should be compared to in PCR validation.
Thanks again - really appreciate your advice!