Hello,
I have been using dada2 to produce a count table from the environmental sequencing we did on soil samples.
As one of the most recommended ways to normalize dada2 produced count table is variance stabilization transformation (vst), I have used DESeq2 package in R and then phyloseq to generate ordination plots.
The procedure is well documented in various pipelines online but I could not really get a clear explanation of the transformed values in a new count table.
First of all, some of the transformed values appear negative. One phyloseq thread explained this as normal and suggested removing negative values from the table. Secondly, I struggle with an intuitive understanding of this table and further processing in phyloseq. tax_glom() function produces very different results for taxonomy ranks that it feels almost impossible to compare between samples in my study (making all of them look very very similar). Thirdly, looking at the bar-plots built on the normalized count table, I do not really know how to understand the difference in height of the bars. They do not correlate with the number of reads per sample and I am not sure how comparable those bars are to each other as the transformation is supposed to be treating the difference of a varying number of reads in each sample.
Have you ever worked with vst on amplicon data? Please share any of your experiences!
Hello, I'm not a daily vst user but I would recommend you to take a look at this paper, it will help you take decision on which method to use.
thank you for your comment. I am familiar with this paper. I would like to understand the vst before rejecting this method and moving to its alternative