ChIP-Seq Simulation

[golharam]

PeakSeq enables systematic scoring of ChIP-seq experiments relative to controls
Nature Biotechnology, Vol. 27, No. 1. (04 January 2009), pp. 66-75.

Modeling ChIP Sequencing In Silico with Applications
PLoS Computational Biology, Vol. 4, No. 8. (22 August 2008), e1000158.

Design and analysis of ChIP-seq experiments for DNA-binding proteins
Nature Biotechnology, Vol. 26, No. 12. (16 November 2008), pp. 1351-1359.

[Nix]

Hmm,

36bp is sufficient for chIP-seq, shoot for >10-20 million duplicate free uniquely mapped reads (1/2 - 1 lane on a GAIIx), and at minimum provide 5ng of correctly sized chIP product to your sequencing core. Lastly, run 3-4 biological replicas for each sample. Barcoding will cut the cost. Very shortly, reviewers are going to reject papers that lack these controls. See http://useq.sourceforge.net/usage.html for additional details.

[dawe]

Hi there!

Does anyone simulate a ChIP-Seq experiment prior to actually running the experiment?

AFAIK no, but if it was possible I would like to know :-)

1) What is the optimal read length to sequence to. Is 35, 50, 75, or 100bp enough? If 50 or 75 is enough to uniquely place 90% of the reads and 100bp doesn't give us any added benefit, then we would only like to sequence up to 75bp and save money.

We currently use 36 bp in single read for human chip-seq and we are happy with it. The IP material is a small fraction of the entire genome and it is highly covered with short reads. You have a 35 Mb genome, which can be covered ~20x with a single average Illumina lane (36bp, SR), your IP will probably be much less. You should be able to have a good IP profile.

2) How much starting material do we need. Is this really something we can answer prior to sequencing? We have enough based on our core facilities recommendations. Is that good enough?

I believe you should refer to your machine standards.

3) How much coverage are we going to need? We expect this would dictate how much starting material we would need, and how many lanes to run.

As said before, I don't think you'll have any coverage issue here.

Is there anything else or is all of this just overkill? I've read 3 or 4 papers regarding modeling ChIP-Seq in silico, but understanding the underlying code is proving to be difficult, and I'm not convinced its right for us.

I just want to make we are on the right track and would appreciate to hear what others are doing.

Can you post those papers?
BTW, the only problem I can imagine is related to a "low" signal (as you said). If your antibody is not specific, well, it may be hard to distinguish the IP from the Input background (needless to ask:
will you run an Input?). If you don't know about Ab performance, it may be useful to run a third sample with an aspecific Ab (such as IgG or something).

d

[golharam]

Does anyone simulate a ChIP-Seq experiment prior to actually running the experiment? This is coming up because our collaborators want to simulate the experiment to optimize the parameters (aka read length, depth of coverage, etc) to make sure they are doing things correctly prior to sequencing. It makes sense considering we have several options on read length, among other things. So my question goes out to everyone, do you perform a simulation on ChIP-Seq experiments prior to running the experiment? We would like to do so for the following reasons: We are performing this experiment on an organism with a genome of about 35 MB, and the signal in our initial experiment is low, but we have a LOT of coverage, probably more than necessary.

1) What is the optimal read length to sequence to. Is 35, 50, 75, or 100bp enough? If 50 or 75 is enough to uniquely place 90% of the reads and 100bp doesn't give us any added benefit, then we would only like to sequence up to 75bp and save money.

2) How much starting material do we need. Is this really something we can answer prior to sequencing? We have enough based on our core facilities recommendations. Is that good enough?

3) How much coverage are we going to need? We expect this would dictate how much starting material we would need, and how many lanes to run.

Is there anything else or is all of this just overkill? I've read 3 or 4 papers regarding modeling ChIP-Seq in silico, but understanding the underlying code is proving to be difficult, and I'm not convinced its right for us.

I just want to make we are on the right track and would appreciate to hear what others are doing.