Search
Question: diffbind : Is it possible to analyze one sample only ?
0
12 months ago by
ZheFrench10
ZheFrench10 wrote:

I have a pair of chipseq. One is really bad. Mapping failed.

I was wondering if it's possible to analyse only one without the pair ?

Other question :

I tried to do the analyze with both of the samples even is one is not very good.

I did two different things : the first time I set the rule that peaks must be in both replicates...that gaves me 1116 differential peaks.

dba_chip_consensus <- dba.peakset(dba_chip, consensus=c(type),minOverlap=0.99)



Then I apply the following rule : the peaks must be at least in one or the other replicate. It gave me 1000 differential peaks.

dba_chip_consensus <- dba.peakset(dba_chip, consensus=c(type),minOverlap=0.01)

Strange, no ? I was hoping that it gaves me more peaks using one or the other... Is that possible ?

modified 12 months ago by Rory Stark2.5k • written 12 months ago by ZheFrench10
0
12 months ago by
Rory Stark2.5k
CRUK, Cambridge, UK
Rory Stark2.5k wrote:

When you include all the peaks (from either replicate), it is likely that most of the additional peaks are unique to one replicate, so they won't be identified as differential due to high variance (and inadequate power). At the same time, adding these peaks changes the distribution of p-values and hence the calculation of FDR, making it less likely that any specific peak will be below your threshold. So what you are seeing looks right to me!

-Rory