Hello, I want to run ssGSEA on my samples, on log2 trasnformed TPM. I read that GSVA recommends to pre-filter lowly expressed genes similar to how one would do for differential expression. However, I only previously did such pre-filtering on raw counts (the standard, DESeq suggested):

keep <- rowSums(counts(dds) >= 10) >= smallestGroupSize
dds <- dds[keep,]

As I understand, Kallisto-derived TPM are usually imported via tximport and not pre-filtered as results() deals with that internally. I was wondering what would be a sensible cutoff for lowly expressed TPM, equivalent to the counts(dds) >= 10. Or maybe how that cutoff is internally defined while working with kallisto -> DESeq? I basically want to do the same filtering outside of the function.

Thank you.

I've seen in some papers ~1 TPM for saying something in mouse/human etc. was "minimally expressed".

For an average length gene, 10 counts out of 10e6 reads would be ~1 TPM, so it seems to correspond also to region above clear shot noise.