distribution of Pvalues in RNASeq data

[biofreak]

Hi All,
I used DEseq package to find differentially expressed genes for my RNAseq data ( 2 conditions 3 samples each). I the plotted all the P values and the distribution is as follows (see attached).
Is the data reliable?
thanks.

[Simon Anders]

If you have two conditions, how come you have plots for 4 groups. (I consider "condition" and "group" as synonyms, but it seems you mean something else.)

Apart from that: Your p values have a strong skew to the right. This typically either means that the variance was overestimated or that there are hidden confounders (e.g. batch effects). Your p values tend to be too high, i.e., the analysis is a bit too conservative, but the hits you get should be reliable.

[biofreak]

Hi Simon,
thanks a lot for the reply.
I am sorry for the confusion.. I have 12 samples (4 groups with 3 replicates each). and I was comparing 2 groups each time.. So is there any way to not over estimate the variance?
Speaking of batch effect, attached is the read quality plots for these 12 samples (4 groups).
I am not sure if this is a batch effect. Could you please help.

[Simon Anders]

No, batch effect means that some samples are more similar than other because they were prepared in the same batch. This can cause a lot of problems. Make a sample distance heatmap as described in the vignette to investigate.

In general, you will have to use the variance as is. DESeq sometimes is a bit too careful, but a loss of power is certainly less of an issue than too many false positives.