I have a large data set of several hundred samples from three locations, male and female from two different species, and I am running ANCOM for all possible location/sex/species pairwise comparisons. The ANCOM results for most of these comparisons look fine, i.e. taxa with large values for “w” marked as “TRUE” for rejecting the null hypothesis show up in qiime2view as significantly different abundances–just what you’d expect. For a few pairwise comparisons, however, all taxa, even those with w = 0, are shown as “TRUE” for rejecting the null hypothesis and appear in qiime2view as significantly different abundances.
Does anyone have any idea what might cause taxa in a comparison with w = 0 to show up as rejecting the null hypothesis? Does this mean the ANCOM went awry or is this a glitch in qiime2view? If there were p-values I feel like I could answer this, but there are not. I am happy to share a “good” and “bad” ANCOM qzv if that would help.
I knew not to trust the significant taxa that were True with w = 0, but there are a few others that are harder to judge. For example, for most of my pairwise comparisons the significantly different abundant taxa have w = ~600, but for the pairwise comparisons with the false signifcant w = 0 taxa I also have a few taxa that are in the range of w = 150. This is lower than the ~600 I see in other comparisons, but obviously also much higher than the clearly incorrect results where w is in the range of 0 to 2.
Any notions on how to evaluate these borderline cases?
Hi @C_Blazier,
You might need to recruit a bit more expert advise but in my opinion the w=150 is high enough to be considered safe. Assuming you don’t have ridiculous # of features that is. Since the W represents the # of significant subhypothesis I would say this is probably safe. You can always validate your findings with another differential abundance tool, or you can do some post-hoc comparisons of those taxa to be sure. I always like to plot the relative abundances of the identified features as well just for my own sanity.
In short, I don’t think W=150 is borderline and those are probably real differences.