What was the model used for the PERMANOVA (e.g. free permutations of residuals - n=5 does not allow for many permutations at the lowest level)?

Dear all,
I used qiime2 to get a pcoa file of Bray-Curtis dissimilarity matrix and the significant difference from PERMANOVA . But when asked "What was the model used for the PERMANOVA (e.g. free permutations of residuals - n=5 does not allow for many permutations at the lowest level)?" what shoud I reply?
My study has two factor A and B. A has 4 level, B has 3 level, a total of 3*4=12 group. Each group has 5 5samples (replicates). Is it not appropriate to ues PERMANOVA?

1 Like

Hi @YuZhang,

This sounds like a kind of technical question. Let me refer you to the original permanova paper by Anderson, which will hopefully help answer your more math-y questions.

With your current sample size and an unblocked design, there are 5^12 possible permutations, or about 244 x 10^{9}. So, there's lots of space to explore. (I think vanilla permanova is an unblocked single variable design; Adonis offers more options for blocks and grouping). However, 5 samples per group may be small to detect a signal. It will likely lead to problems with other tests (ANCOM will struggle for instance), so just be aware.

Best,
Justine

3 Likes

Thanks, it seems like,I should change the permanova to the Adonis. But, how can I sove the problem

Hi @YuZhang,

This is more a note of caution about your design and being careful with your beta diversity.

There isn't a good solution for this problem after samples have been collected and sequenced. The best time time to solve this is during the experimental design phase. The tests carry the sample size limitations they have and you're ultimately paying an FDR penalty if you're doing untargeted testing.

If you run into the isssue when you get there, here are a few threads on the subject

1 Like

This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.