By looking at the alpha-rarefaction plot alpha-rarefaction.qzv, using the sample metadata column set as TRT, it seems that there is a significant difference between EXCENEL and NO_TRT treatments, judging by the the box plots across sequencing depths for both metrics - Faith's PD or Shannon.
However, when testing Faith's PD differences between groups faith-pd-group-significance.qzv (selecting column = TRT) there is no significant difference. How come there is such a difference in the alpha-rarefaction plot but none in the subsequent analysis?
Thanks for sharing your QZV files, this made investigation pretty simple. First, I opened up faith-pd-group-significance.qzv and determined the rarefaction depth you chose:
Okay, so it looks like you selected 22456 as the rarefaction depth. So now if I look up that depth on the alpha-rarefaction.qzv file:
The results actually appear to agree really well at that rarefaction depth.
Some things to keep in mind - you have effectively removed all but ~5 samples from your study by selecting such a deep depth. You will probably need to select a lower sampling depth, that still takes into account the breadth of samples in your study.
Thanks for helping me troubleshoot this. As you noted, I have selected 22456. This actually only excludes 2 samples from the analysis. The remaining 103 samples are maintained.
The picture that you are showing is on a depth which is ~10x greater than the depth I have selected.
Please look again at the alpha-rarefaction plot. You will note that at this depth, there seem to be a difference between treatments (please see the attached screenshot).
Hi @JoaoGabrielMoraes, thanks for pointing that out, yeah, I totally misread the x-axis and was off by quite a bit! Okay, so the other thing to point out here is the box-and-whiskers in the two plots are showing different distributions. In the alpha rarefaction plot, it is the distribution of all samples in the metadata group, from all of the rarefaction trials (10 by default, I think). In the alpha group-significance plot it is only looking at the samples in each metadata group for a single rarefaction trial. The reason alpha rarefaction runs multiple trials is to give you a sense of how "stable" the alpha diversity is at a given depth, in addition to how it changes at different depths. Make sense? You could try re-running alpha-rarefaction, setting the number of trials down to 1 and comparing there (don't forget though, the sampling is random, so will be different each time).
