Statistics with Rarefaction

[Samuel Major]

Hi Qiimers,

I have some concerns about the effects of rarefaction and which type of rarefaction to use when performing things such as ANOSIM and then SIMPER to explain ANOSIM results.

Initially, I did the above analyses on some data without rarefying (it looks like there's been some debate about whether rarefying should be used to calculate statistics) and saw that ANOSIM told me my samples were significantly different from each other. Now I'm curious to know if this is due to the size of each samples and I'm getting these types of results because of biases without sub-sampling. 

So now, I'm trying to sub-sample so I can compare each sample when they all have the same number of OTUs to get a better representation of why or how they're different from each other. I saw another poster had concerns about performing the single rarefaction command because the sub-sampling method may take a bunch of rare OTUs for the sub sample. 

My question is which would be the best rarefaction technique for getting the best ANOSIM or ADONIS reults; single rarefaction or multiple rarefaction? Also, can someone explain the difference between multiple_rarefaction.py and mulitple_rarefaction_at_even_depth.py?

Thanks

Sam

[Se Jin Song]

Hello Sam,

Yes, there is certainly some debate over to rarefy or not, and the answer comes down to the specific analysis you're performing.  If you have doubts, it's good to check using both to make sure your results don't change. 

Single rarefaction vs multiple rarefaction is a matter of how much you think each rarefaction will differ from the next. Generally, the deeper you rarefy, the smaller the effect of sampling rare OTUs will be, so if you're rarefying at a shallow level, your results may differ more between rarefactions than between deeper rarefactions. Again, one way to check this is to perform multiple rarefactions and run ANOSIM or ADONIS on each rarefaction to see how results change/don't change.

mulitple_rarefaction_at_even_depth.py performs a user defined number of rarefactions, each at the depth of your choice, but only at one depth (e.g. 10 different rarefactions, each at 5000 sequences per sample).  multiple_rarefaction.py on the other hand, does the same thing, except for different depths at increments of your choice (e.g. 10 different rarefactions of 1000 seqs/sample, 10 rarefactions at 2000 seqs/sample, etc...)

Hope this helps!

Se Jin