meaning of a peak's p-value?

[Hershel Safer]

Hi Tao,

I would like to clarify what is being measured by a peak's p-value. I am guessing that it is the probability of seeing a fold-change this high or higher in a null model that has peaks appearing in some random fashion along the genome, but I would like to be able to say more exactly. How would you characterize it? Thank you,

Hershel

[Tao Liu]

Hi Hershel,

For each peak region, MACS calculates the a local lambda for poisson
distribution based on the control tags within the 1kb, 5kb and
10kb(1/5/10k are parameters that you can modify) nearby regions to
consider the local fluctuations and biases. The local lambda is the
maximum of the averages of tags for 1/5/10 kb regions and a whole
genome background. Then this local lambda is used to calculate the p-
value of poisson distribution. If there is no control data, the ChIP
data will be used instead, where the 1kb region is not considered. The
fold-enrichment is also calculated using local lambda.

Hope it helps,
Tao

[Hershel Safer]

Thanks for the explanation.
Hershel

[jane]

Hi Tao,
I have tried different lambdasets,and the results are quit
different,so how to modify the lambdasets to match the CHIP data
best(I have no control tags)?

Regards,
Jane

[Tao Liu]

Hi Jane,

On Aug 9, 2008, at 11:42 PM, jane wrote:

> I have tried different lambdasets,and the results are quit
> different,so how to modify the lambdasets to match the CHIP data
> best(I have no control tags)?

To change the parameter for lambdaset is not recommended. However, if
users really want to play with this parameter, they should keep in
mind that, the three regions in lambdaset (default 1k, 5k, 10k) is to
consider a most nearby region, a modest big region and a large region
to find the local bias around the peak. A reasonable set must not be
too similar, too small, or too large. The default works well in our
test suites -- chIP-seq for human CTCF, NRSF and FoxA1. You may need
to find a optimal value for your own chIP-seq, but in fact, tweaking
this parameter should not affect good peak which has low FDR, big fold-
enrichment, and high '-10*log(10,pvalue)', if the parameter is
reasonable.

Regards,
Tao

[jane]

Thanks! I will take your advise.

Regards,
Jane