Which DE method perform better for small number of genes

Question

[Thind amarinder]

I would like to know which differential expression method performs better for genes (miR) number around 500 and sample number around 27. I have DE results of small-RNAseq based on negative binomial (Deseq2, EdgeR), poison (NoiSeq) distribution and t-test. It would be great if someone explains how gene number can effect DE.

[ATpoint]

Fyi NOIseq will be deprecated with the next release of Bioconductor, makes it questionable if you want to build on it.

Answer 1

[Devon Ryan]

Gene number has little bearing on which test to use, the answer is always DESeq2/edgeR/limma-voom unless you have hundreds of samples.

[Thind amarinder]

Thank you for your answer. May I ask, what if I have both samples and variables(genes) are of similar range.

[Devon Ryan]

I guess if you're doing a targeted panel then that might be the case. You're still better off with DESeq2/edgeR/limma-voom.

[Thind amarinder]

I found an explanation from https://dx.doi.org/10.1016%2Fj.ajog.2006.07.001 when the number of measurements (arrays) is small two methods for gene selection using a public dataset: fold change and a moderated t-test.