diffReps - ChIP-seq differential analysis package

[asiangg]

Dear colleagues,

I'd like to introduce you to diffReps -- a ChIP-seq differential analysis package that my group developed.

A platoon of peak calling programs have been developed in the past few years to identify TF binding sites and chromatin modification sites under a single condition. However, the methods that are available to identify the differences between two conditions are surprisingly scarce. This is an important question because ChIP-seq is now routinely applied to in vivo tissues where two or more states need to be compared: be it disease vs. normal or laboratory treatment vs. control.

This kind of problems has its unique challenges. Changes are often subtle and do not stand out as clear as peaks vs. background. Biological replicates are commonly used to increase the statistical power. It is critical to select the appropriate statistical test for this task. It is also crucial to handle the large ChIP-seq files as painless as possible.

In the past 3-4 years, my group have handled hundres of ChIP-seq samples, the majority of which come from in vivo mouse brain. To handle this large amount of data, we decided to develop our own program because none of the method on the market appears to be satisfactory to our needs. The result is diffReps -- a program package that integrates data manipulation, statistical tests, downstream annotations, etc., all into one command line. It literally saved our lives and tons of time.

diffReps is developed in PERL and it runs on all platforms such as Linux, Mac and Windows. It uses only modest RAM and finishes running in reasonable time, even for large files. It also integrates two very useful tools. One is called region analysis, a program that can annotate differential sites or peaks into genes or heterochromatic regions. The other is called hotspot finding, a program that can identify locations where the differential chromatin modifications happen more often than random.

diffReps is shared as an open source project and can be found here:

https://code.google.com/p/diffreps/.

I'm also happy to announce that the manuscript about diffReps has been published by PLOS ONE and can be located here:

http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0065598

If you are working on ChIP-seq data and would like to compare two conditions, I suggest you to give diffReps a try. There is also a discussion group about diffReps, where you can post your usage questions or receive announcements:

https://groups.google.com/forum/?fromgroups#!forum/diffreps-discuss

As we are still working to improve diffReps further, please sign up to become a member of the discussion group so that any future releases and news will be directed to your inbox.

I hope you find diffReps to be useful to your research.

[biznatch]

Your PLOS ONE paper only looks at histone modifications. Would diffReps be appropriate to use with transcription factors that don't generate wide peaks like histone modifications? Perhaps using a smaller window size and step size would be better in this situation?

Would you recommend removing duplicate reads (ie. PCR duplicates) before running diffReps?

Is the strand (+/-) required in the .bed file used for input, or just the first 3 columns (chromosome, chromStart, chromEnd)? I have paired end reads so I was thinking I could combine them, since it doesn't look like diffReps can take paired end reads. If I do this then strand is irrelevant, so I could leave it blank or just say "+" for everything.

[asiangg]

biznatch,

Yes, you can certainly use diffReps on transcription factor ChIP-seq. I don't see any reason why it cannot be used on TF. Using smaller window size is recommended for TF data.

And I would recommend removing PCR duplicates before diffReps, which provides a more conservative list.

If you have paired-end data, you can manipulate them as you have already done and set the fragment size to 0. The strand info is used to shift short reads. Once you set the fragment size to 0, they become irrelevant.