Understanding differential abundance analysis (gneiss)

Hello,
I’m new in qiime2. Firstly, I want to thank all of you for make qiime so accesible to non-bioinformaticians like me! I already completed the “moving pictures tutorial”, and I’m trying to following the gneiss tutorial. I want to ask to something related to the latter, specifically regarding balance proportion plot interpretation. I followed other issues opened related to gneiss but not specifically to the following question:
For instance, if I found a relevant balance (100 taxa in the numerator/15 taxa in the denominator) related to one variable (such as type of treatment= A or B) and I plotted it. Graphically, It seems that treatment B are displaced to the left of the plot, which means more presence of the denominator taxa in such group.
a) Could I infer that ALL taxa included in the denominator of the balance (n=15) are, in general, related to treatment B, or only such taxa included in the proportion plot (n=10) must be considered as significant?
b) And if not, how could I investigate if the 5 taxa included in the denominator but not in the proportion plot are also related to treatment B?

Thanks a lot and, again, congrats!

This is very good question. The devil is in the details. Gneiss is testing for the log means between these two groups - so you can imagine scenarios where the numerator is largely described by 1 taxa (in which case none of the other taxa are significant).

That being said, this depends on the tree - and the variance-based clustering should ward against these sorts of issues.

It all boils down to how you determine your reference microbes. Gneiss tries to automatically find a references to draw comparisons.

If you are interested in microbes in the denominator - it makes sense to try to test to see if the difference between log-mean of those microbe vs the log-mean of the numerator microbes changes substantially across the conditions. There are no established qiime2 commands that can currently handle this - but there is software in the works that can handle these fine-tuned questions (see: https://github.com/fedarko/rankratioviz and https://github.com/biocore/songbird)

3 Likes

OK, thanks for your helpful answers!

Regards

This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.