Dear all, we run DESeq2 using the standard approach described in the manual, including filtering through the rowsums function, starting from RSEM-generated counts . We ended up with several statistically significant genes that had very low TPM counts overall, and reasonable TPM only on few samples. In order to takle this issue, we were thinking of using the edgeR function filterbyexprs to keep only genes that are expressed over 10 counts in a number of samples at least equal to the size of the smaller comparison class (or fractions of it). The approach would be used just to filter out low-expressed genes, than the remaining raw counts would be processed with DESeq2 under the standard approach. Would this be fine for DESeq2? Thank you for helping.

Yes, filterByExprs is fine to run upstream of DESeq2().

Sometimes I recommend the following type of code on the support site:

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