Pseudocounts in RNA-seq analysis

Question

[Marco Pannone]

Hey everybody!

I would like to hear your opinion about using pseudocounts when performing RNA-seq analysis. I have always been following the standard protocol for DESeq2 when analyzing RNA-seq data, using their great guide, but recently a colleague pointed out that I should always add the value 1 to counts of all genes. He mentioned that this will help for those genes which have a count of 0, in order to avoid dealing with negative values after transformation.

I always filter out low counts genes with:

filtered <- rowSums(counts(dds)) >= 10

So I wonder if DESeq2 already deals with pseudocounts by default or if it's required that I perform this step myself.

Thanks!

Answer 1

[ATpoint]

Pseudocounts are commonly added before log-transformation of normalized counts for downstream analysis, to avoid taking logs of zeros and logs of values > 0 < 1 which as you say would be neagtive. For the DE testing you should not do that. Give it the raw counts as clearly described in the manual.

[Marco Pannone]

Thanks you for your answer! So for example I should use pseudo counts when plotting heatmaps, volcano plots, PCA plots, etc.? And when it comes to GSEA, where I use as input the normalised read counts matrix for all genes and replicates, should I also use pseudocounts in that case?

[ATpoint]

I have not used pcs for anything but preventing taking logs of zeros.