Hello,
Even after data transformation (microbiome data of taxa abundance) my data looks like negative binomial distribution instead of Normal distribution. I used log transfomation of the data.
Is it OK to perform ordination and further statistical analysis with in R with this data or there is something else ?
Hi Prashant,
I would say it probably depends of the questions you want to ask. Could you provide us with additional details, on whether you want to compare the relative abundance of specific taxa between two groups, if you want to show that the overall taxa abundance distribution between two groups changes, etc.
There are some useful pointers that do the statistical analysis using the Negative Binomial nature of your distribution, or a Dirichlet-multinomial as in Article Hypothesis Testing and Power Calculations for Taxonomic-Base...
Cheers,
Kevin Vervier
Many thanks Kévin Vervier!!
I need to test each of the species in my data for their differential abundance between groups and select few according to the p-values.
Thanks for the paper as well, I would try to read and implement those ideas if possible.
You're welcome.
The problem I usually face in this case is that the test run on most abundant species (lets say Wilcoxon-Mann-Whitney) is not suitable for rare species, especially if there is a lot of zero abundance (problem of zero-inflated data well known in single-cell RNA-seq).
I guess it is the same with Microbiome data. A lot of zeros and that make data transformation and normalization quite tedious.
If you suspected your data are zero-inflated (check the overdispersion), you might want to perform adequate statistical tests relying on this assumption. Here is a nice review that will give you the right pointers depending of the scenario; Article Hypothesis Testing and Statistical Analysis of Microbiome
Thanks Kévin Vervier .. I will find time to read it fully but I can see this paper mentioned the CSS normalization. CSS worked quite well for my 16S data. One other method of the inverse hyperbolic sine transformation for data normalization
log(yi+(yi^2+1)^1/2).. also worked well for ITS data.
- Badges
- Science topic
- Similar topics
- Methodology
- Research Methodology
- Research Papers