Thank you very much for all the informative discussion, I'm learning a lot from your conversation.
Today, I want to ask about interpretation, and how to report output of differential abundance analysis conducted by gneiss.
I performed correlation-clustering, ILR transform and linear regression for 3 treatment conditions adjusted for age, sex, BMI (two categories) as formula "~AGE+SEX+BMI+TREATMENT" and got output. I got a question regarding interpretation of coefficient p-value table.
So, I want to report whether abundance are different in treatment condition.
If I want to report these results, Can I say that (of course with considering the numerator/denominator balance) taxa contained in balance y0 are differently abundant by age, and taxa contained in y1 are differently abundant in treatment condition 2? (As p-value other than treatment condition 2 are not statistically significant in y1.)
What can I report about the taxa contained in y2?
Hi @heimer. Awesome! Glad you got some potentially interesting output.
Concerning how to interpret the results, pvalues are great, but note that they only give you one bit of picture. If you have a small effect size (i.e. small coefficients), chances are it isn't a very meaningful result.
Given the results that could be potentially interesting, I'd try to get some visualization to directly observe the effect size. I'd first look at the plots in the balance-taxonomy command to sanity check the effect size for the balances that you find interesting. If there isn't much overlap in the boxplots / scatterplots, it could be an interesting result.
However, your case is a little more complex, since you are trying to control for the treatments given the age and sex. After running the balance-taxonomy command, it may be worthwhile to directly unpack the balances from the resulting qza file from the ILR transform, and plot the interesting balances separately in Python/R, so that you can try to control for these variables. For instance, it maybe worthwhile having a scatter plot of balance vs age, and plot the treatment groups separately to see how much separation there is between the treatments.
