My previous microbiome statistical analysis has been based on normalized order-level metagenomic read counts of samples from two experiment groups. After confirming that the samples overlapped in the multivariate space and that the dispersion was similar ("betadisper" from vegan package, not significant), I performed PERMANOVA ("adonis" (vegan)) to look for differences between the two groups.
In my current experiment, I work with the same type of data, but from multiple experimental groups. One of the groups don't overlap with the others in the multivariate space.
My questions are the following:
1) A key assumption of PERMANOVA is similar multivariate dispersion within each group. Is overlap of groups in multivariate space equally important?
2) Is it sufficient to perform an overall test of betadispersion (including all groups)?
3) What if one of the groups have a different dispersion or do not overlap with the other groups in multivariate space? Can I still use PERMANOVA including this group?
4) In general, after running an overall PERMANOVA with all groups, can I use PERMANOVA for a posteriori testing of each pair of groups?
Even Sannes Riiser,
University of Oslo, Norway