maxit parameter cannot be changed in DESeq function

[nalcala]

Hi,

When launching the DESeq function, I get the following warning:

25 rows did not converge in beta, labelled in mcols(object)$betaConv. Use larger maxit argument with nbinomWaldTest

Nevertheless, there is no way to specify the maxit argument to the DESeq function so it can pass it along to nbinomWaldTest. Could this feature be added?

Thanks in advance!

Nicolas

[mikelove]

hi Nicolas,

You can just instead run nbinomWaldTest() directly after and this is equivalent to re-running DESeq() with more iterations. You can also substitute DESeq() from the beginning with the equivalent code:

dds <- estimateSizeFactors(dds)
dds <- estimateDispersions(dds)
dds <- nbinomWaldTest(dds)

For further questions please post to https://support.bioconductor.org

[nalcala]

Hi Mike,

Thanks for the prompt answer. Yes, this will solve the issue. Nevertheless, I still think that adding maxit as a parameter would be beneficial to the users, unless you want to drop the function DESeq in the future.

Best,

Nicolas

[mikelove]

Noted.

[gabrielet]

i have a follow up question, regarding the set up of the maxit parameter. what if, when using the methodology outlined above i am unable to reach convergence using a maxit=10000000? I thought I may increase the betaTol threshold, i.e. from the default 1e-8 to 1e-7 but I am unsure about how this parameter is influencing the computation and how reliable the estimates may be after the tweak. any hints on this?

[mikelove]

Usually this occurs when the model is not a good match to the data. When running DESeq2 on typical bulk RNA-seq it converges with a few iterations, e.g. often <10 or so. What kind of data do you have?

[gabrielet]

yep, i am not analysing RNA-seq data. i am using DESeq2 to run some differential abundance on bacterial amplicon data, i.e. ASVs matrix, so apparently that is the problem.

[mikelove]

Yeah, i'm skeptical of the applicability of DESeq2 here.

[gabrielet]

yes, I read some papers stating that DESeq2 is not a good fit for such data but I wanted to try it out in combination with other tools. however, if you say it's really not a good idea, then i will stick to other methods that are explicitly made for this purpose. thank you very much for your time!