ANCOM giving me W scores of 0

I ran ANCOM on a set of data that had min samples of 2 and a min frequency of 10 (recommended from the “ANCOM giving strange W values” thread) and I’m still getting W scores of 0. Should I be taking these low scored taxa as significantly different in my analysis?

You mean that these taxa are detected to be significant even with W=0? That’s a bit funky, could you link your qzv, and qzas with the script to generate the qzv?

I'm attaching the finalized ANCOM level 2 and the file used to create it (file was located in a day_1_oral file).

qiime feature-table filter-features \
    --i-table day_1_oral/filtered_table.qza \
    --p-min-frequency 10 \
    --p-min-samples 2 \
    --output-dir day_1_oral-ANCOM
qiime taxa collapse \
    --i-table day_1_oral-ANCOM/filtered_table.qza \
    --i-taxonomy taxonomy.qza \
    --p-level 2 \
    --o-collapsed-table day_1_oral-level_2
qiime composition add-pseudocount \
    --i-table day_1_oral-level_2.qza \
    --o-composition-table day_1_oral-level_2_with_pseudocount.qza
qiime composition ancom \
    --i-table day_1_oral-level_2_with_pseudocount.qza \
    --m-metadata-file merge_no_blank6.tsv \
    --m-metadata-column delivery_mode \
    --o-visualization day_1_oral-level_2_with_pseudocount_delivery_mode.qzv

filtered_table.qza (686.2 KB)
day_1_oral_filtered_features-level_2_with_pseudocount_delivery_mode.qzv (975.6 KB)

Ok, so not all of your W scores are 0, so that’s promising

Note that you only have 10 phyla, so you would expect your W scores to be low – the largest W score you can have is 9.

Have you considered running this at genera or sequence level? ANCOM is more advantageous for larger datasets

Yes I’ve run them for both just so I can compare at multiple levels. Do you recommend a score at which the comparisons become statistically significant? Or should I only be ignoring a score of 0? Thank you!

ANCOM should automatically recommend where to accept the null hypothesis or not (you should be able to download those results in the csv file at the top).

However, it looks like ANCOM is recommending to reject the null hypothesis for all of the genera - this is weird behavior, I wouldn’t trust these results at the phylum level and consider those at the genera level.

This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.