Unifrac or Martin's P test? We did Illumina MiSeq on the V4 region and used QIIME for processing the data. Now we want to statistically compare a handful of microbial communities that clustered together in a PCoA plot. In a pairwise comparison of the five samples, I got extremely different p values: Unifrac ~ 1 and P test
Would you be able to use R? Then I will recommend MetagenomeSeq a package in R made by Bioconductor. Let me know if you need help...
I would need more detail about the main question of your study to give a concrete answer but based on what you described above I can say that UniFrac is a good method to determine whether communities are significantly different, comparing many communities simultaneously using clustering techniques such as PCA.
Attached you will find an excellent paper about the multivariate analysis that you can apply if your are focused on ecological aspects.
Jacob, thanks for the MetagenomeSeq suggestion. I haven't used that, but will look into it...eventually! I'll reach out if I need help when I do.
Nataly, we've done the clustering, via PCoA, using both phylogenetically-based (Unifrac) and non-phylogenetically based dissimilatory metrics. Some communities cluster together, and we're interested in being able to say "these groups are NOT statistically different." (We're deeming that a worthwhile statement. Whether it is [or not] would be an interesting conversation.) Thanks for the paper. It is a good one, and one that I should be referencing right now! Unfortunately, it doesn't tackle Unifrac or P tests for statistically comparing communities.
Dear Joe,
As far as I can understand, you want to compare groups of microbial communities on the basis of multivariate descriptors. If you have a hypothesis about groups (such as different treatments) then I suggest using MANOVA with permutation testing, which allows different choices regarding resemblance functions between the communities. This will provide you with a P-value, which will allow judging your hypothesis. MANOVA with permutation testing is described in Pillar & Orlóci (1996, see http://ecoqua.ecologia.ufrgs.br/arquivos/Reprints&Manuscripts/Pillar_Orloci_1996_JVS.pdf). See also Anderson (2001, Austral Ecology 26: 32-46). However, if your question is about clustering, then a test with MANOVA will be circular, because the groups will be defined by the same data used for the test. In this case I suggest using a bootstrap resampling method, which is described in Pillar (1999, How sharp are classifications? Ecology 80: 2508-2516. available at http://ecoqua.ecologia.ufrgs.br/arquivos/Reprints&Manuscripts/Pillar_1999_Ecology.pdf). These methods are implemented in MULTIV, which is available at http://ecoqua.ecologia.ufrgs.br/ecoqua/MULTIV.html.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis