I reviewed the QIIME2 doc and other posts about whether ANCOM could be used for multiple group comparisons. I found the answers to be pretty controversial.
I'd like to compare the specific taxa differences between four groups and I was able to run the ANCOM through QIIME2 without any errors. Would that be the correct way to do so? or I will need to compare two groups at once?
Also, if I have to compare two groups at once, which one could be the reference group? For example, I'm comparing Group A to Group B, and Group B is the reference group. I'd like to know which taxa are more abundant in which group compared to the other group.
Ancom-BC as implemented in Qiime2 can be used for multiple groups comparison to the reference. That means that all groups will be compared to one group that was selected as reference (or the first one in alphabetic order if not provided).
If your goal is to compare each group VS each other, or pairwise comparisons, then you need to check Ancom-BC 2 in R, which can perform such comparisons but not yet added to Qiime2. Please note that running Ancom-BC in Qiime2 for each pair of groups to perform pairwise comparisons, though can be technically implemented, is not a substitute for pairwise comparisons between multiple groups since Ancom-BC FDR correction performed by features, and can't be reproduced just by pooling all the p-values with following correction.
Ancom-BC results can be visualised by DA-barplot (check docs, available plugins), with positive LFC (to the right) for features more abundant in tested group and negative LFC (to the left) - in reference.
Also, all p-values, adjusted p-values, LFC and other info can be exported from Ancom-BC output files.
Please note that running Ancom-BC in Qiime2 for each pair of groups to perform pairwise comparisons, though can be technically implemented, is not a substitute for pairwise comparisons between multiple groups since Ancom-BC FDR correction performed by features, and can't be reproduced just by pooling all the p-values with following correction.
Can I use ANCOM-BC in QIIME2 for pairwise comparison by just changing the reference group? I know it's not technically a "pairwise comparison" but I'd like to compare any of the two groups.
Ancom-BC FDR correction performed by features, and can't be reproduced just by pooling all the p-values with following correction
I also don't quite understand about the stats you mentioned here. Could you elaborate it for me?
When performing pairwise or multiple comparisons, it is important to adjust obtained p-values for False Discovery Rate, or FDR. As far as I know, in qiime2 Ancom-BC performs FDR corrections within the features, so it is not so easy to readjust "raw" p-values from the output.
By doing so, you will increase the amount of comparisons you are performing, so FDR correction is not more valid. And as I wrote above, it will be not so easy to readjust p-values for FDR, since your comparisons are based on 2 or more separate runs of Ancom-BC.
It is why I referred to the Ancom-BC 2 in R, which can perform pairwise comparisons in one run and will adjust p-values for FDR under the hood for all comparisons performed.