ANCOM rejects null hypothesis at many W values that are 0

Hello all,

My ANCOM results show that it rejects null hypothesis at many W values that are zero.
I'm not sure if this is a bug or something else?
(I read somewhere else in this forum that this is not supposed to happen)

Please kindly find attached the results:
ANCOM_level2.qzv (427.7 KB)
ANCOM_level6.qzv (448.4 KB)

Also, a higher W value means that it's more statistically significant, am I right?
But does this mean that I can compare W values at different levels?
For example, if taxa A at level 2 has W=20, while taxa B at level 6 has W=15, does it mean that taxa A is more statistically significant than taxa B?

Basically, does it behave like p-value?

Many thanks for your generous time and help

Hi @fgara,

There’s been a fair bit of discussion among the mods of feature or bug with regard to having all 0 W values. (I’m in the “feature” camp, BTW). ANCOM tests to be quite conservative, and so my first three questions for you are

  • Did you see a difference in beta diveristy between your groups in any metrics? (But specifically in an abundance-based metric)?
  • How many samples do you have total and per-group?
  • How did you/did you pre-filter your data to boost your power?

Okay, this is another case where you probably want to look at the effect compared to the W. W measures how often the taxa is statistically significant based on your “significance” criteria. It doesn’t tell you about the significance. So, I don’t think that’s a great interpretation.
You may instead want to look at something like the test statistic or some other comparison to get a better sense of how big the difference is.

Finally, because it will. bug me for the rest of the day (and possibly into my weekend) if I dont say this - Id discourage the phylum level testing. If there’s something interesting going on, you’ll find it at a lower taxonomic level with good filtering. Plus, I’m not sure you buy yourself that much more information about your ecosystem with such a broad categorization, but that’s my two cents, and ymmv.



Hi Justine,

Thank you so much for your reply!

I will check these and get back to you.

Ah I see, thank you! I will look into that.

This is very interesting… thanks a lot, I didn’t expect that but I’m very interested to learn more about this. Perhaps you have papers or other examples that I can learn from? We did see some interesting insight at the phylum level from a different study (same data sets but different grouping / metadata feature).

Many thanks Justine, I’m always really thankful for your valuable insights!

Hi @fgara,

Glad it helped!

Some of this is anecdotal. I suspect that if you go back to that earlier analysis and look, you’ll find a set of specific features that drive that phylum level difference. (It’s also worth keeping in mind that taxonomy is nested.). I dont think its a great example of a paper or how to do analysis, but if you look at Table 3 of this study, and you look up the taxonomy associated with each level, you’ll find that it’s nested and the difference in Fusobacteria is driven by the difference in Leptotrichia. (However, I dont recommend using the paper as an example for much else, I dont think they did a great job with respect to that nesting aspect.)

I think there’s also an ecological question - is it important to know that global warming is associated with a change in the number of things with spinal cords :woman_scientist::shark::dog2::bird::frog: and the number of things with shells :lobster::spider::bug:, or is it more helpful to know a specific set of species within phylum chordata is driving CO2 and CH4 production :woman_scientist: ::cow2:? I think community level data is really important, don’t get me wrong, but Im not sure that collapsing to that high of level works as well as alternative statistics (alpha & beta diversity).


1 Like

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