Is the only difference between vst and varianceStabilizingTransform the first step of estimation, in which, vst uses a subset of genes while varianceStabilizingTransform uses all info? Therefore, vst is a faster method than varianceStabilizingTransform.

When the number of genes/ features is low and using the default vst parameters gives an error message of

"less than 'nsub' rows with mean normalized count > 5"

I wonder whether executing the following will essentially equalize vst and varianceStabilizingTransform:

# vsd <- varianceStabilizingTransform(dds) # is this the same as the following?
nsub_input <- sum( rowMeans( counts(dds, normalized=TRUE)) > 5 )
vsd <- vst(dds, blind=TRUE, nsub = min(nsub_input, 1000)) ## I found that blind = TRUE is the default of varianceStabilizingTransform

I am curious whether using varianceStabilizingTransform preserves more information from the data (i.e. that is more "accurate") in the case of there are more than 1000 genes for subsetting in vst while I can tolerate a slower calculation (as long as it can be finished in a reasonable time). I know vst is robust, but would like to know what would be the best practice. I also read the discussion here, but would like to confirm my understanding is correct. Thanks!

I had answered previously, on Biostars:

https://www.biostars.org/p/459013/#459016

Kevin