Normalization of ASV table: rarefaction or variance stabilizing transformation (VST)

Hi all,

Does anyone use variance stabilizing transformation (VST) for the normalization the ASV table, which is recommended when using DeSeq2 for microbiome data analysis?

Some people don’t like rarefaction as they think rarefaction compromise with samples having relatively low sequencing depth and sacrifice reads.

What’s your opinion on these two randomization methods?

1 Like

Hi @Claire010,

That is the recommended normalization method for DeSeq2. Differential abundance methods in QIIME 2, such as ANCOM or Aldex2, have their own normalization strategies that are automatically run (and should not be used on rarefied data). Various published benchmarks with these methods have demonstrated superior performance to DeSeq2 for microbiome differential abundance testing. Hence, we recommend other normalization strategies and differential abundance testing methods, and do not have a DeSeq2 plugin in QIIME 2.

Agreed… for differential abundance testing, this is well accepted at this point that rarefaction is not an appropriate normalization method. For alpha and beta diversity testing, on the other hand, the results are more mixed, and rarefaction is often still performed as an easy and historically well-accepted method for sampling depth normalization (as some normalization is required). Several other QIIME 2 plugins, such as DEICODE and q2-breakaway, are available and offer alternative approaches for alpha and beta diversity analyses without rarefaction.

So to summarize: different methods used in QIIME 2 (and beyond) employ their own normalization strategies, especially differential abundance techniques (none of which use rarefied data). Only a few standard diversity methods in QIIME 2 employ rarefaction, but alternatives to these exist for those who want to use other normalization strategies.

There has been more discussion of these in the “general discussion” category on this forum, I recommend searching around for those to find more existing discourse on the topic.

2 Likes