Dear all, I have found this paper: piRNAs Are Associated with Diverse Transgenerational Effects on Gene and Transposon Expression in a Hybrid Dysgenic Syndrome of D . virilis

in which there is this statement: "We did not perform DESeq2 analysis because estimation of the dispersion parameter with DEseq2 is unlikely to be robust with about 200 TEs, compared to standard mRNA-seq analysis that estimates the dispersion parameter using thousands of genes".

How many genes is advisable to use with DESeq2?

Thank you.

Riccardo

It's hard to say exactly because it depends on the heterogeneity of the genes. If they all have similar true dispersion values, then seeing a few is sufficient. If they have very different true dispersion values, and these have strong dependence on the mean, then we need to see more genes to have a good estimate for the dispersion prior. I would guess (thinking about the steps involved: fitting the parametric or loess curve to the gene-wise estimates) that less than 100 is too few and more than this is probably sufficient. If I had 100-200 genes and wanted to use DESeq2 I would use the fitType="mean" instead of the parametric or loess curve.

Thank you.

Riccardo