W too high and interpretating results from collapsed/feature table

Hello everyone,

I'm actually running and trying to interpret ANCOM results. Briefly some important stuff about my data:

I'm working with microbiota from mosquitos midgut, I have 90 samples, I filtered my samples with a frequency like the depth chosed from the rarefaction that I used in diversity analysis (3800). Furthermore I filtered features with less than 10 reads and removed singletons.

Here are my filtered table and the filtered collapsed table (level 6): filtered-new-table-min10-samp2.qzv (455.8 KB) filtered_table_level-6.qzv (475.5 KB)

So I runned ANCOM wiith one category from my metada ("local") and in this category I have 7 groups. I have more three categories, but I'm trying first to understand my preliminary results until step forward.

I've runned ANCOM with my filtered table and with the collapsed table (level 6) to compare results, for my surprising I have in both ANCOM volcano plot a too high W. I compared with others results from the Moving Pictures, Parkison's tutorial and in the forum. My W is very high and in my x-axis (F-score) I do not have negative numbers.

I don't know if this occurs normally, the ANCOM results from the collapsed table have a slightly smaller W.

Here I will post my code, and the files of the results from the filtered and collapsed-filtered table. But my questions are: There are some bias in my results? I can confirm which taxa (or features) are differentially abundant using the TSV table below as a parameter? Why my W it's too high?

qiime feature-table filter-samples \
  --i-table filtered-table-trim-gg-decon.qza \
  --p-min-frequency 3800 \
  --o-filtered-table filtered-new-table.qza

qiime feature-table filter-features \
  --i-table filtered-new-table.qza \
  --p-min-frequency 10 \
  --p-min-samples 2 \
  --o-filtered-table filtered-new-table-min10-samp2.qza

qiime composition add-pseudocount \
    --i-table filtered-new-table-min10-samp2.qza \
    --o-composition-table filtered-new-table-min10-samp2_with_pseudocount.qza

qiime composition ancom \
    --i-table filtered-new-table-min10-samp2_with_pseudocount.qza \
    --m-metadata-file Metadados.txt \
    --m-metadata-column local \
    --o-visualization ancom_local.qzv

ancom_local.qzv (450.7 KB)
ANCOM - local results.tsv (25.7 KB)

qiime taxa collapse \
    --i-table filtered-new-table-min10-samp2.qza \
    --i-taxonomy taxonomy-gg.qza \
    --p-level 6 \
    --o-collapsed-table filtered_table_level-6.qza

qiime composition add-pseudocount \
    --i-table filtered_table_level-6.qza \
    --o-composition-table filtered_table_level-6_with_pseudocount.qza

qiime composition ancom \
    --i-table filtered_table_level-6_with_pseudocount.qza \
    --m-metadata-file Metadados.txt \
    --m-metadata-column local \
    --o-visualization ancom_collapsed_local.qzv

ancom_collapsed_local.qzv (447.9 KB)
ANCOM collapsed - local results.tsv (20.7 KB)

Hi happy new year to everyone,

Updating: I tried with other categories from my metadata, with collapsed and the feature table and the results are quite similar with a high W and a F-score positive.

I'm still stucked in the interpretation of this results

Hi João and happy 2022!

To detect likely differences in abundance, ANCOM is testing for significant differences in ASV ratios between groups. For each ASV, the number of "successful" tests like that is called W. This means that the maximum W equals the number of ASVs in your table minus 1, in your case 612 - 1 = 611 for the ASV table and 203 - 1 = 202 for the collapsed table. This also means that W values higher than in the tutorial are nothing to worry about and, in fact, you have taxa in your data set with significant differential abundance (even reaching maximum W).

Regarding the x-axis, I only see negative and positive values (the classical volcano crater shape) when I compare exactly two groups. If I compare more than two groups, all clr values are positive, which is also what you see, so I think it is normal. I think the values are made positive during the calculation, but someone else may know why exactly that is.

You may find the following threads useful:

I hope that cleared some of your concerns!

2 Likes

Thanks a lot @mverce !

That's pretty clear now, so grateful for your help.