I’m using PERMDISP to evaluate if some treatment groups cluster more closely together than others in my weighted unifrac beta-diversity analysis, based on distances of samples from centroid in each group. However, when I download the TSV file from the PERMDISP results, I see that PERMDISP results are based on the distances among all samples in each group and not in distances of all samples from the group’s centroid. I’m wondering if there is a way to do PERMDISP based on distances from centroids.
Thanks
Thaisa
Welcome to the forum @tcantuju!
The data download shows the raw data used in those plots and those pairwise sample distances do not reflect what this test uses.
It looks like the PERMDISP test that is run here uses the spatial median rather than the centroid (and not the pairwise distances):
http://scikit-bio.org/docs/0.5.4/generated/generated/skbio.stats.distance.permdisp.html
Would that work for you? If it must be centroid you will need to run this in another method, since that option is not exposed in QIIME 2.
Good luck!
I found that there is a stronger interference of outliers when spatial distances are used and that is why I wanted to use distances from centroids. I could exclude outliers for the analysis (because now I found they actually don’t meet my inclusion criteria), but then treatments will have a different number of samples and that is not good when using PERMDISP, right?
It is too bad that I cannot run distances from centroids on QIIME though
2 Likes
This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.