deseq2 pseudo counts
2
0
Entering edit mode
Last seen 6.8 years ago
United States

For each sample (and replicate), do I have to convert zero read counts to ones (pseudo counts) while running deseq2?

I assume I don’t have to, but a quick clarification.

deseq2 • 1.7k views
0
Entering edit mode
@mikelove
Last seen 44 minutes ago
United States
Hi Prasad, No you don't have to convert. The zeros (including when all samples have a zero), are automatically handled by the model (and by the software).
0
Entering edit mode

Hello, Michael!

How exactly does DESeq2 handles with the zero counts? I tried to understand and look for it but is still unclear for me. Thanks

0
Entering edit mode

Here is the DESeq2 paper with details on the methods:

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4302049/

Counts of zero are not treated differently by the statistical model than a count of 1, 2, ...

0
Entering edit mode

Hi Michael! But when I try to use only the normalized data from DESeq2, I can see that the data [gene] with at least one value as "zero" is eliminated. This will be affecting the result - that is in this case, when such genes are eliminated it's not possible for us to know the gene's expression value in other tissues.

1
Entering edit mode

If you mean it is filtered, that is controlled by independentFiltering in results(), just set it to FALSE.

In this case, I would recommend a minimal filter at the top of your script, before DESeq():

keep <- rowSums(counts(dds) >= 10) >= X
dds <- dds[keep,]


Some recommendation for X is the sample size of the smallest condition group (this from limma/edgeR software guides).

0
Entering edit mode

Hi Micheal!

Thank you for the quick response.. Firstly, I congratulate and Thank you and your team for your efforts on DESeq2.

Secondly, I fixed the problem by adding your suggestion. I just tried to change it to zero.

keep <- rowSums(counts(dds)) >=0
#keep only these data
dds <- dds[keep,]


Now, I have the genes which had one or two zero values with other integer values with its normalized counts- ie., zero counts are not discarded.

0
Entering edit mode
Last seen 6.8 years ago
United States

Thank you very much.