normalizing relative abundance biom OTU table

[Vanessa V.]

Hi all, 

We would like to run statistics from the relative abundance biom OTU table and want to normalize the relative abundance to run statistics.  First, is this possible to do with relative abundance data (or do you have to normalize from the raw OTU table with sequence counts) and if so, is there an easy way to do this through QIIME?  What method can be used?  I was thinking we could normalize the relative abundance on a scale from 0-1 across samples but I wasn't sure what others have used.  

Thanks for your help in advance, 

Vanessa

[Jamie Morton]

Hi Vanessa,

That is actually an interesting question.  

The normalization technique most commonly used in qiime is rarefaction.  This technique has shown to reduce some of the bias associated with low coverage.  This is particularly important when you are using presence/absence metrics such as unweighted unifrac or binary jaccard distance.

However, there has been some controversy surrounding this approach, especially in the context of differential abundance (i.e. determining what taxa are different between samples),

and numerous alternative normalization methods have been proposed.  We have implemented a few of these normalization methods here.

Long story short, if you are running distance based statistics (i.e. compare_categories.py), it is probably safer to rarefy.  But if you are interested in differential abundance, I'd check out the normalization and differential abundance scripts.  It is also important to note that there will be more statistical techniques will be become available in QIIME2.

Hope this helps!

Jamie