RNA-Seq replicates with big differences in medians of VST values

Question

[panconchoclo]

Hi,

I have RNA-Seq data (5 biological replicates) and after the Variance Stabilizing Transformation (VST) normalization I found that few replicates have the medians really low regarding the other samples/replicates (see plot). I used blind dispersion estimation in DESeq2. I did the same in other tissue and this is not happening and I used the same treatments for both. Based on PCA analysis is possible consider some of these replicates with low median values like "outliers" as well but not all of them. So I would ask you if a kind of threshold index exist to use that to select replicates which should be removed before perform a DE analysis and thus avoid the bias produced by the replicates with low medians. I remember that for DE analysis with microarrays I saw a index like that but I have never seen something like that in DE analysis with RNA-Seq data.

Thank you very much for your comments,

Answer 1

[shunyip]

Those "low" medians are zero. You should be able to correct it by filtering low count genes. I suggest only using genes that have an average of at least one count in those samples that have the median close to zero.

[panconchoclo]

Thank you so much shunyip for your answer! Filter out the genes with few counts fix the low medians itself in the box plot but just now I realize that remove those genes is unnecessary for the DE analysis because DESeq do that by default by “independent filtering”. https://support.bioconductor.org/p/63234/