diff_analysis with multiple groups

[dalferes]

Hi,

I was hoping to get a bit more understand of the diff_analysis, if possible. I am struggling to understand why I get different taxa as significant from the diff_analysis function if I compare my 4 groups vs if I subset the data and compare 2 of the groups.

I have a class group with 4 groups - A, B, C, D. When I use diff_analysis I get only 3 discriminative features after lda.

If I subset the class group to only have options A, B - when I use diff_analysis I get 78 features discriminative features after lda.

The code used was:

set.seed(50)
deres <- diff_analysis(obj = ps_sub_ran_for_tree, classgroup = "cluster",
mlfun = "lda",
filtermod = "pvalue",
firstcomfun = "kruskal.test",
firstalpha = 0.05,
strictmod = TRUE,
secondcomfun = "wilcox.test",
subclmin = 10,
subclwilc = TRUE,
secondalpha = 0.05,
lda=2)

The results when comparing just A and B match a lot of the ones found using random forest, while comparing A,B,C,D does not. Should I be doing each pairwise comparison separately?

Thank you for you help!