Ancom error, reporting all taxa as significant

Dear Qiime2 developers,

I am trying to understand which taxa differ between the 2 experimental groups at four different intestinal regions, so I have split my dataset by intestinal regions, removed all 1-sample ASVs and run ANCOM (at ASV level and collapsed L2-L7 levels).

For certain intestinal regions, ANCOM outputs ALL taxa as significant, but with very low W values (see Duodenum L6 attached table_filtered_MC_Duodenum_only_filter2_collapsed_L6_CFTR_genotype.qzv (346.3 KB)
). However, in other intestinal regions, I see a more normal output of just a few significant taxa.

I suspect that when ANCOM outputs all taxa they are all false-positives and there are actually no significant taxa in this region, do you think this is correct?

I just want to make sure I am safe using the output from the other intestinal regions where there are only a few significant taxa identified.

Thank you for your help,

I think that is more weird behavior going on with ANCOM - if all of the taxa are significant, ANCOM definitely will not be able to detect that.

This may be more of a question to ask either Siddhartha Mandal or Shyamal D Peddada as listed on the ANCOM paper

1 Like

Hi @JenKelly, just wondering if you resolved this issue or perhaps got more information from the authors as @mortonjt recommended? I ask because I’m reaching almost the identical problem myself and wanted to avoid contacting the authors if this has already been looked into.

Some info about my data specifically:
I am comparing 2 groups which are actually the same samples pre and post treatment. I know there are more appropriate methods for comparing paired microbiome data but this a bit exploratory and I’m using it to compare to other methods.
I’m expecting very little to no changes between the 2 time points so the ANCOM assumptions of less 25% changing taxa should be ok.
The output has lots of low or zero W values that are being identified as significant. The table itself has been collapsed down to the genus level and I’ve tried various parameters to remove rare taxa and noise. I’ve tried the dataset on with as low as 33 features up to 77 and basically the same issue arises.
Sampling depth is good, ranging from ~10k-50k. The n values per group are not great with 7 and 8 samples, but using another subset of the data with similar n values, treatments, and conditions gives me 0 or 1 significant taxa, which is what I expect.
Looking at the raw data itself I’m willing to accept that certain taxa are actually different across the groups but those taxa are among those that are showing as W=1/0 and significant, so its hard to trust what is happening.

Any guesses as what is driving this is welcome!

1 Like

Hi, I was struggling with the same situation. For me worked filtering table to remove features that only occurs in 3 or less samples. After it my ANCOM output was showing significant features with high W.

Hi @timanix,
Thanks for your input! Removing rare taxa to reduce noise is usually my first recommendation in this scenario as well. But in this case I’ve tried various parameters to achieve this, including removing any features not occuring in at least 4,6,8 samples and min-frequency of 10-50-200. They all produce the same issue, which is why I’m struggling with this… :thinking:

1 Like

I was thinking recently, what if one of features is present in samples from one time point and completely missing in another, which outcome it will be for this samples? If will not be filtered out by searching for rare features since it can be found in many samples from the same time point. Is it still will have strong W or not?

1 Like

Of possible interest to this topic, see this post regarding the low W values being considered differentially abundant.

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