Hello everyone, I work with a ChIP-seq experiment profiling a histone modification in wild type and a mutant. In the mutant we expect a general reduction of the modification. It is possible that in some regions the reduction is more pronounced than in others. We want to test 2 hypotheses: (i) modification levels are lower in the mutant than in the WT. (ii) Reductions in modification levels differ between genomic contexts. My questions is how to best normalize such data between samples. I am afraid scaling (e.g. based on library size) or quantile normalisation will greatly reduce or even mask the differences between WT and mutant (imagine all peaks are reduced by a factor of 2). Is there a robust method for such cases? Maybe based on a statistical modelling of unspecific background vs. specific signals, where only the latter should be affected in the mutant? Thank you in advance.

The best way to do this is certainly by spiking in external controls, there was a recent Genome Research paper on this. It is hard to do what you suggest, because in my experience the proportion of reads coming from the specific signal (the ChIP enrichment) differs a lot from sample to sample and simply reflect the overall efficiency of the specific IP reaction. I don't believe you can expect the same enrichment efficiency in every experiment. In my experience, analyzing a similar experiment, I was able to get something reasonable by doing a simple scaling, despite the clear expectation that it might not work ... I agree that if all peaks were reduced by exactly 2-fold you would not see anything, but in practice that did not turn out to be the case in our experiment, despite a expected global effect. In the future we will probably do the external spike- in. In the absence of that, you may have expected "household" peaks, although I suspect not. Best, Kasper On Mon, Jun 2, 2014 at 9:58 AM, Minerva Trejo <minerva.trejo@slu.se> wrote: > Hello everyone, > > I work with a ChIP-seq experiment profiling a histone modification in wild > type and a mutant. In the mutant we expect a general reduction of the > modification. It is possible that in some regions the reduction is more > pronounced than in others. We want to test 2 hypotheses: (i) modification > levels are lower in the mutant than in the WT. (ii) Reductions in > modification levels differ between genomic contexts. > My questions is how to best normalize such data between samples. I am > afraid scaling (e.g. based on library size) or quantile normalisation will > greatly reduce or even mask the differences between WT and mutant (imagine > all peaks are reduced by a factor of 2). Is there a robust method for such > cases? Maybe based on a statistical modelling of unspecific background vs. > specific signals, where only the latter should be affected in the mutant? > > Thank you in advance. > _______________________________________________ > Bioconductor mailing list > Bioconductor@r-project.org > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: > http://news.gmane.org/gmane.science.biology.informatics.conductor > [[alternative HTML version deleted]]