How to get normalized count table from DESeq?

Question

[leranwangcs]

Hi,

I'm using Deseq compare differential abundance. Here is my code:

ds.all <- phyloseq_to_deseq2(ps0.infant.pbs, ~ sample_type)

geoMeans <- apply(counts(ds.all),1,gm_mean)

ds.all <- estimateSizeFactors(ds.all,geoMeans = geoMeans)

dds.all <- DESeq(ds.all,fitType = "local")

Then as the results I got 8 ASVs that showed significantly different.

My questions:

1. I used geoMeans is because otherwise DESeq() would fail with error:

    Error in estimateSizeFactorsForMatrix(counts(object), locfunc = locfunc,  : 
   

   every gene contains at least one zero, cannot compute log geometric means

   So does that mean these two steps:

   geoMeans <- apply(counts(ds.all),1,gm_mean) 

   ds.all <- estimateSizeFactors(ds.all,geoMeans = geoMeans)

are normalization steps of DESeq? Or there are other hidden normalization steps in DESeq?

1. How can I extract the normalized count table that DESeq used to generate the 8 ASVs? I tried counts(ds.all), but the count table is exactly the same with the raw count table.

2. I also tried counts(dds, normalized=T), it does look like a normalized count table, but how can I know if this is the exact normalized count table that deseq used for its analysis?

Thanks! Leran

[swbarnes2]

I take it that for this application, it's okay for every gene to have a zero? Because in the regular bulk seq RNA that DESeq was designed for, that's not typical unless you have a couple of failed samples included.

[ATpoint]

phyloSeq manual instructs to do exactly that so... ¯_(ツ)_/¯

Answer 1

[ancient_learner]

something like this should work:

 Normalized <-counts(dds.all, normalized=TRUE)

[leranwangcs]

Thanks! I know that there are multiple normalization methods contained in DESeq, how can I know if Normalized <-counts(dds.all, normalized=TRUE) gives me the exact same count table that DESeq used for DE? And what does the geoMeans do here?

Thanks!

[ATpoint]

The geometric Means are an attempt to dampen the effect of outliers/genes that are different between samples in order to find a factor that balances the two samples at a neutral baseline. See for details:+