I am new to running ANCOM and in my analysis I am comparing 2 groups of participants gut microbiome. I ran ANCOM and the results showed all w=0. I filtered minimum frequency to 10 as a previous post suggested.
Welcome to the forum!
Could you clarify what the issue is that you are looking to resolve?
At what taxonomic level are you operating at? How many features are in your table? Have you seen this post also for a similar inquiry, can you try the advise there and let us know if you have additional questions.
Thanks, I really appreciate this forum, and have found many answers in my journey.
I forgot to mention in my previous post that the groups differed in alpha (evenness, but not others) and beta-diversity (weighted UniFrac, but not unweighted UniFrac and BC). The ANCOM analysis at level 2-6 showed all W=0, after I filtered out low frequency feature ( p-min-frequency 10, p-min-samples 2)
I am attaching some of the results here. I wonder if this due to the distribution of the features, it seems that there are many low frequency features. Thanks!
Thanks for providing us the files, those are very useful. It is a bit strange to me that you see W scores of 0 right across the board. Do you see a similar pattern if you were to perform ANCOM without collapsing your data at a taxa level (i.e using the ASV table). At L6 level you only have 64 features, so the max W score you can get would be 63, which maybe a bit too low for ANCOM to perform appropriately, and this would only get worse as you use lower level grouping like Order, or phylum. However I’ve personally seen ANCOM perform fine with 60-100 features so there is also the possibility that this is simply the nature of your data. How many samples in each of the groups you are testing? I couldn’t find this info from the provenance.
One group has 13 people and the other has 21 people
I re-ran the analysis with the entire ASV table (1142 features), it came out the same, all w=0;
I also went ahead and ran L6 without any filtering (total ~300 features), all w=0 again.
In this case, does it mean ANCOM may not be appropriate for our data? what is an alternative?
Thanks for the update. Yeah this is not really an expected behavior, at least not from any data set I’ve seen. I can ask around a bit to see what the ANCOM experts think, but for the time being I would recommend not relying on these results and instead try your luck with q2-aldex2 or q2-songbird.
Will update here if I can!
Thanks for the suggestions.
I tried q2-aldex2, and nothing came out significantly different, at different levels.
I have not tried songbird, it seems that I will need to play with different parameters. will dig a bit more.
If aldex2 also does not show any significant features then it is likely that you simply don’t have any differential abundant taxa. You may be under-powered to detect a small effect size as well. One recommendation that @jwdebelius made, which I think would be a good sanity check, is to use a more stringent filtering step to reduce the penalty of FDR corrections.
And yes, songbird does take some parameter tuning but it is a very powerful tool once you work out its ins and outs.
@anitalijia, do you have an observed difference in beta diversity? If not, then not having an difference in differential abundance is expected.
If there is a difference in beta diversity observed, then perhaps you’ll want to try aldex or songbird as @Mehrbod_Estaki suggested.
This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.