Question about weird ANCOM results

[Manuss_Ponce (Manuel Ponce Alonso)]

Hi everyone!

I'm doing a 16s metagenomic analysis in Qiime2, and I have obtained surprising results from a ANCOM analysis, which I could't explain: My feature-table includes data from stool and sputum samples. When I performed the ANCOM analysis in sputum data (comparing treatment vs. placebo) at phylum level, all phyla showed differences in abundance between treatment and placebo (but with low values for w and f in the vulcano plot). However, when I performed the same analysis in stool data, no by-phylum differences were encountered. I'm not sure if those results are related to errors from the data or incorrect approach. I attach here some results and the pipeline used. Thanks in advance!

Pipeline (for stool data):

qiime feature-table filter-samples --i-table results-dada2/tablef.qza --m-metadata-file metadata.tsv --p-where "Sample='stool'" --o-filtered-table differential-abundance/stool-tablef.qza

qiime taxa collapse --i-table differential-abundance/stool-tablef.qza --i-taxonomy taxon/taxonomy-gg.qza --p-level 2 --o-collapsed-table differential-abundance/stool-tablef-phylum.qza

qiime composition add-pseudocount --i-table differential-abundance/stool-tablef-phylum.qza --o-composition-table differential-abundance/comp-stool-tablef-phylum.qza

qiime composition ancom --i-table differential-abundance/comp-stool-tablef-phylum.qza --m-metadata-file metadata.tsv --m-metadata-column Treatment --o-visualization differential-abundance/l6-ancom-Subject-stoolf.qzv

metadata.tsv (5.5 KB)
taxonomy-gg.qza (82.1 KB)
l6-ancom-Subject-sputumf.qzv (331.6 KB)
l6-ancom-Subject-stoolf.qzv (330.9 KB)
tablef.qza (125.5 KB)

[thermokarst]

Hey there @Manuss_Ponce!

@Manuss_Ponce

I’m not sure if those results are related to errors from the data or incorrect approach

Why do you think there is an error here? Sputum vs stool seem like they originate from pretty different environments to me - I am not surprised that you are observing these kinds of differences between them. I would recommend taking a look at q2-gneiss for some additional differential abundance tests you can run to provide additional insight into your data.

Hope that helps!

[Manuss_Ponce (Manuel Ponce Alonso)]

Hello Thermokarst, I followed your advise and tried to run my data using gneiss, and I think I understand better my results. Thanks!

[mortonjt (Jamie Morton)]

Hi @Manuss_Ponce, one thing worth keeping in mind is that you shouldn’t expect compositional methods to be consistent between phyla and OTUs. This is because the collapse taxa module either computes sums or means, rather than products or geometric means.

If you really wanted consistent phyla / OTU results, we’ll need to enable collapse-taxonomy to allow for this. These problems will be apparent in both ANCOM and Gneiss.