Dear EdgeR users

I am working on a RNASeq dataset with extremely small sample size (2 samples in each group; N=4) and I am trying to find any DEG's using these samples (I am aware of the consequences of using very small sample size). I did find some DEG's with P.values as low as 6.9e-17 and an equivalent FDR of 8.14e-13. Although it seems encouraging to see some differences in my samples, I am more concerned about so low P.values with such a small sample size (This is because of my limited statistical knowledge). Just wondering how EdgeR calculates P.values with such low sample sizes. Any comments or opinions are greatly appreciated. Many thanks for your help.

Best

Venu

The calculation depends on the type of analysis you performed with edgeR. P-values are calculated using the exact test for classic edgeR, the likelihood ratio test for GLMs, and the quasi-likelihood F-test for QL-based analyses. If you want more details on how each of these work, have a look at the corresponding papers that are referenced in ?exactTest, ?glmLRT or ?glmQLFTest.

Generally speaking, getting very low p-values with small sample sizes is not an issue. If you have counts of 1000 in one group and 0 in the other group, then obviously you'll get very low p-values. Remember that edgeR also shares information across genes to improve the stability of the dispersion estimates. This mitigates the loss of power due to small numbers of replicates in typical RNA-seq experimental designs.