Search
Question: use only lib size as normalizing factor in edgeR
0
26 days ago by
Udi Landau30
Udi Landau30 wrote:

Hello,

I would like to use only the lib size as normalizing factor in edgeR, so I am doing this:

cds <- DGEList(counts = counts, group = group, norm.factors = MyNormFactors)

I skip the

cds <- calcNormFactors(cds, method = "none")

Because It will bring the norm factor back to 1..

and I continue with calculating dispersion and doing the exact test.

What is seems to me strange is that the pseudo count look the same as the original count.

Isn't it supposed to change?

When I use no normalization (factor = 1)  the pseudo count do change a little..

Just want to know if this is the right way so I could trust my results..

modified 25 days ago by Aaron Lun21k • written 26 days ago by Udi Landau30
2
25 days ago by
Aaron Lun21k
Cambridge, United Kingdom
Aaron Lun21k wrote:

There's several issues in your post.

• If you want to do library size normalization, then don't specify norm.factors in the DGEList call. Having all normalization factors equal to 1 means that edgeR will use the log-library sizes directly as offsets, without calculating effective library sizes.
• It doesn't matter if you skip calcNormFactors or not, your normalization factors should already be 1.
• All currently recommended edgeR pipelines use glmQLFit and glmQLFTest instead of exactTest. The quasi-likelihood GLM framework is more accurate, more flexible and faster than "classic" edgeR - better in every way.

Thank you!

Can you tell me what would you recommend to use for plotting the data?

I understand that the pseudo count is not suitable?

1

See Section 2.15 of the edgeR User's Guide ("Clustering, heatmaps etc").