High Variation Within Biological Replicates For Significant Genes Detected By Deseq/Edger

Question

[Gangcai]

Dear all, I have used both of edgeR and DESeq to get significant differentially expressed genes. Both of them output some candidates which have quite a lot of variation within biological replicates. such as:

t1  t2    c1     c2
0    24    0    0
2    22    0    0

correspondent pvalue(adjusted) is 0.04.

   t1/t2: biological replicates for treatment 
   c1/c2: biological replicates for control

Is it common to have such kind of low p value by using negative binomial test?

Thanks in adance~

[Michael]

I would think that is normal, what is your question? It depends on the number of replicates and the variation in general.

[Vitis]

At least for edgeR, I think it's a modified exact test, which may not use the variations among replicates in the same way as those tests for continuous distributions.

[Jeremy Leipzig]

I would alter the numbers so they read like 10 24 0 0 or 20 22 0 0 just so see what pval for a "non-dubious" gene would be