ChIP-Seq: knockout as control?

[Combo]

I have four samples.

1) A standard ChIP for a transcription factor (TF) of interest.
2) A standard Input DNA control (i.e. sheared, but no IP).

3,4) Same as 1&2 but from Knockout cells completely lacking the TF.


What's the best way to utilize these controls?

I have been running peak callers using the Input DNA as control (i.e. 1 vs 2, and 3 vs 4). Then I subtract the overlapping peaks found in the knockout peak calling data from those in the wild-type.

Would it be better to run the peak analysis using just 1 vs 3 (i.e. wild-type ChIP vs knockout ChIP)? What's the ideal way to do this?

Thanks for any ideas

[kopi-o]

There is probably no "right answer" but your approach seems good to me. It seems more rigorous than ChIP-seq analyses usually are.

[HESmith]

If the input samples 2 and 4 are identical, then you can directly compare experimental samples 1 and 3. If not, then the approach you're using is better than direct comparison.

[Combo]

Thanks for the responses.

At this stage I'm trying to make the best of the data. It definitely has problems -- e.g. even after lots of MACS parameter optimization, a 'known' enrichment area only shows up in the peaks results with an FDR ~20%. I don't think I've seen any papers using such a high cut off.

Any thoughts on this or what changes could be made to improve it? I've tried many combinations of 'keepdup' and 'mfold', which have helped, but I'm running out of ideas now as far as making this validated enrichment area show up with a low FDR.

I will try the 1 vs 3 comparison later today. Maybe that will help.

[Combo]

Interestingly enough, performing MACS peak calling with the knockout pulldown as control (1 vs 3) improved the FDR of the peak within the validated/known occupancy region considerably.

Perhaps I will next run MACS on the knockout vs knockout input, then subtract those peaks from the above.