I have a couple questions for a project I'm working on.

First, I originally used the DESeq2 tool on a local instance of Galaxy (I find the GUI really easy to use, plus I can upload batches of featureCounts files rather than combining them manually). Today, I went to check my results by running it in R. Both were fresh installs, so they should be the newest version on their respective platforms (I do not know, however, if they are the same version between both).

I was somewhat surprised to see that the results were not the same. They were close, but not the same. There were different P values, fold changes, orders of genes, etc. The base means, however, seem to be the same. Why would this be? Are they simply different versions? I know the methods have changed a bit over the years, but I'm not sure when they changed most recently.

My second question involves the alpha setting. I know that ideally, alpha should be set ahead of time to whatever Padj. cutoff I want to use. However, Galaxy doesn't seem to have the ability to change the alpha value (except for use in the MA plot). When I got my original results from the Galaxy tool, I used a Padj. cutoff of .05, I suppose out of habit. I know that generally the automatic value for alpha is .1, and I assume the Galaxy tool uses this. So, it seems I may have used the default alpha of .1 but then a cutoff of .05.

How much of a problem is this? When working in R today, I ran the res() function with both alpha = .1 and alpha = .05 and the resulting tables were identical. I totally get that setting alpha to your planned cutoff is ideal, but is it necessary? Should I be rerunning all of my analyses, or can I still generally trust my results? Explain this to me as if I haven't taken statistics in years... because it's true.

Thanks so much in advance for any advice!

DESeq2 methods have been pretty steady since 1.16, when lfcShrink was introduced. This was Bioconductor release 3.5 in 2017 where we are currently on Bioc 3.17.

https://github.com/mikelove/DESeq2/blob/devel/NEWS

If you can find out the DESeq2 version you were running within Galaxy that would give a clue as to the difference.

Setting alpha in results has very little impact typically so I wouldn't worry about re-running analyses. alpha tells the independent filtering procedure what target FDR threshold to optimize for, when choosing the minimum average count for filtering. Therefore it also has no effect if independentFiltering=FALSE and will generally have low or not effect for many well powered datasets.