FrederickHuangLin / ANCOMBC

Differential abundance (DA) and correlation analyses for microbial absolute abundance data
https://www.nature.com/articles/s41467-020-17041-7
108 stars 29 forks source link

Dunnett test is more stringent than pairwise test #276

Open LLansing opened 4 months ago

LLansing commented 4 months ago

I've ran the same ancombc2 analysis as both Dunnett and pairwise, but the Dunnett results have fewer significantly differential taxa. My understanding was that the Dunnett test employs fewer comparisons and thus p-value adjustment would make a smaller impact compared to the pairwise results. However, it appears I may be mistaken.

Is my understanding incorrect?

Here are my function calls:

 da_sp <- ancombc2(data = species_pseq,
                 fix_formula = "trt.foster+rep",
                 p_adj_method = "BH",
                 n_cl = 4,
                 pseudo_sens = TRUE,
                 group = "trt.foster",
                 pairwise = TRUE)

  da_sp_dun <- ancombc2(data = species_pseq,
                    fix_formula = "trt.foster+rep",
                    p_adj_method = "BH",
                    n_cl = 4,
                    pseudo_sens = TRUE,
                    group = "trt.foster",
                    dunnet = TRUE)

to which I then compare the number of values that are TRUE in da_sp$res_pair$diff<comparison> columns to da_sp_dun$res_dunn$diff<comparison> columns.

Maggie8888 commented 2 months ago

Dunnett test controls multiple comparisons, you may find helpful information form the website https://en.wikipedia.org/wiki/Dunnett%27s_test