Ancom-bc2 and use of structural zeros

General Discussion
1

Hi everyone,

I have a question regarding the use of --p-group and --p-structural-zeros in qiime composition ancombc2 .

I am analyzing an in vivo experiment where one microorganism was intentionally administered to treatment groups. As expected, the control group has 0 reads for this microorganism across all samples, while the treatment groups show presence.

When I run ANCOM-BC2 without --p-structural-zeros , the microorganism appears as differentially abundant. However, when I include:

--p-group Group
--p-structural-zeros

the microorganism no longer appears as differentially abundant.

My understanding is that ANCOM-BC2 classifies it as a structural zero in the control group. However, I am unsure how this should be interpreted biologically and statistically:

  • Is the taxon still considered differentially abundant but treated differently by the model?
  • Is it expected behavior that a taxon completely absent in one group does not appear in the standard differential abundance output?
  • In this type of experimental design (intentional inoculation), is it recommended to use --p-structural-zeros , or could it mask biologically meaningful differences?

Thanks in advance

2

Hello,

Ancom-BC2 tries to find features that are differentially abundant, and "--p-structural-zeros" option removes features with 0 counts in all samples from one group, meaning that it will be missing in the output.

No, it is filtered out from the DA test

With enabled "structural zeros" it is expected.

I would run both to see whether other taxa disappear from the output together with the supplemented taxon, since this may indicate a connection or contamination. Then, I would use the results from the run with enabled structural zeros for the DA report and supplement it with a presence/absence matrix (for multiple taxa or groups), or just write it in the text if there are only 2 groups and one disappearing taxa (supplemented one).

2 Likes