I know using Salmon for quantification of gene expression is quite popular while I am still old-school that using featureCounts from the Rsubread package. I am confused that what is the input matrix requirement for both DESeq2 and EdgeR? More specifically, I found that Salmon supply an estimation of gene expression in form of doubles while featureCounts always gives integers. When I was taught in class several years ago that DESeq2 only accepts counts as input, but there are indeed many ways to construct the DESeqDataSet object according to the DESeq2 vignette.
Thus, I wonder if an integer input matrix is a hard requirement for both DESeq2 and EdgeR? I found that in the vignette, the author tried to round the decimals to construct the count matrix (they also stressed that is for demonstration purpose). I would like to confirm if I will need to round up/ down the NumReads column from the Salmon for downstream analysis. Similarly, I would like to confirm that I should not use the TPM column for both DESeq2 and EdgeR.
May I know whether these rules for bulk RNA-seq are still true for analyzing the single-cell RNA-seq data? (I have another question in another post.)