What Parameters Do You Use With Findpeaks For Chip Seq Peak Calling?

Question

[Stew]

I would like to use FindPeaks to call peaks on chIP-seq data, where I have an IP against a sequence specific transcription factor, using Input as control. I would appreciate any guidance as to which control methods, cutoff and FDR methods people are using.

I am currently running MACS and SWEMBL also, and though I get good agreement between these two, I do not get very good agreement with findPeaks.

Specific command line options or general guidance about use would be helpfil. I am using it on pre-filtered (on alignment score) BAM files.

Answer 1

[Dawe]

Hi, I use FindPeaks (FP4) and MACS as well. My typical FP4 parameters are:

-dist_type 1 200
-subpeaks 0.2
-duplicatefilter
-bedgraph
-auto_threshold 0.95

I then filter .peaks file for p ≤ 0.01 (or less) and ht on the auto_threshold results

[Stew]

That is very helpful thanks, not very different to what I as doing. What do you mean by filtering on "ht on the auto_threshold results" though?

[Dawe]

If you run FP4 with -auto_threshold [?] parameter, you'll notice a line in the log file that will tell you the min. height you should filter (per chromosome) to achieve a [?] confidence on your peaks. I've recently defaulted that flag to 0.99

[Stew]

Great, thanks. I didn't know that part. I will take a look.

Answer 2

[Husen]

I have a similar problem with FindPeaks, the bedgraph output is too big compare to output of other peak callers like MACS and Erange.

That is what I use: -aligner bed -dist_type 1 -duplicatefilter -bedgraph

Do you know if there is any option to get less peaks in the bedgraph output?

thanks for your help

[apfejes]

Hi Husen, I'm the author of FindPeaks, so I can tell you that there are a lot of options - but they're not bedgraph specific. The bedgraph option just dumps out the output in bedgraph format. You need to use the program options to filter the output, such as -minimum or even the -control or -compare options.

Answer 3

[suman]

Hi,

I have a question regarding the -iterations and -landerwaterman options.

I read that the landerwaterman options best with dist_type 0 as it performs poorly with weighted (-dist_type 1) distributions and the native/PET (-dist_type 3) distributions.

a) Is the value of integer following dist_type 0 the tag length like 26 reads or 36 reads that we provide in MACS?
b) Can I use both options together in a command?

suman