DESeq2 and R versions for differential_abundance

[Egle Cekanaviciute]

Hi, 

I noticed that the results from using DESeq2 as a method to compare relative abundances in differential_abundance.py depend on the version of R and the version of DESeq2, especially with low sample size.  Specifically, R version 3.1.2 and DESeq2 version lower than v.1.4 give much lower p values/more "significantly different" OTUs than R version 3.2.3 and DESeq2 v1.4. 

This has actually been noticed before by people using DESeq2 within R: 
https://support.bioconductor.org/p/73676/#73694
https://support.bioconductor.org/p/83645/
https://support.bioconductor.org/p/71985/

From these and similar posts it appears that the DESeq2 developers are supporting v1.4 or higher now, which gives higher p values/fewer "significantly different" OTUs. (Which I personally like better too, because I think that DESeq2 has an issue with false positives...)

So it might help to clarify which R and DESeq2 version to install because otherwise the results will be different depending on the version? Also, which versions are canon for QIIME developers?

Another thing that I think might help is that these discrepancies are really strong with low N per group, but at N>=20 per group they basically disappear, so maybe it would be helpful to include a minimal sample number for differential_abundance.py? 

Thank you!

Egle

[Embriette]

Hi Egle,

I've forwarded your message to two experts on DESeq-you should be hearing from them soon!

Embriette