I am having a difficult time understanding what the output files from some of my beta diversity analyses mean (I am using QIIME2 version 2018.6). My metadata includes a categorical variable with two levels: “Yes” (which means the samples were exposed to a certain treatment) and “No” (which means the samples were not exposed to that treatment).
First, I followed the Moving Pictures Tutorial and ran the qiime diversity core-metrics-phylogenetic command to generate the core-phylogeny statistics and the PCoA (with unweighted UniFrac values) file. I then ran the qiime diversity beta-group-significance command to run the PERMANOVA (with unweighted UniFrac values).
My first question is: even though the PERMANOVA results shows significance (p < 0.05), why do the box and whisker plots not appear to reflect this?
I also visualized the PCoA, as mentioned above. Since the PERMANOVA was significant, I expected to see more distinct clusters of the “Yes” and “No” datapoints, but as you can see, there is no apparent clustering. Can you help me understand why these results seem to be contradictory? I know to rely on p-values instead of visual images when assessing significance, but I do not understand why these figures do not seem to support the PERMANOVA results.
quite possibly because the dispersions are different. Use beta-group-significance --method permdisp to test and compare. If permdisp is significant, then it is quite possible that permanova is only reporting a significant result because of different dispersions between groups.
permanova and PCoA do not always correspond perfectly if you have weak separation of groups… the main reason being that with PCoA is a method of dimension reduction and you are looking at the top 2-3 axes (most likely), while separation may be more apparent on a different axis. permanova is looking at pairwise differences overall and hence does not relate perfectly to pcoa visualization. In your case, my money is on dispersion though, not on separation along a different axis.