Another question: based on the gene coordinates in my bacterial gff, ~30% of them have overlaps with other genes. With the pipeline's mapping and counting approach, bowtie2 only reports the best alignment for each read, so a read that is mapped to the overlapped region between two genes is forced to be mapped to only one of them. Salmon can deal with multimapped reads through its EM algorithm, but it doesn't take any effect if the input BAM has no multimapping. Is my understanding correct? Based on your experience, how impactful is this issue? Would it be better to let bowtie2 report multimapped alignments so that Salmon can make use of them?

bowtie

does not care about genes. Is it mapping against the genome or the transcriptome? If the former, there is no multimapping. If the later, in my experience these overlaps are small (but I don't know what bacterium you are working with), so only a few reads will be ambiguous (i.e. not enough to change your results in a meaningful way, no matter what happens). I agree, though, that multimapping and splitting the counts by Salmon would be preferable here. (The overlaps are in operons and are a few bases, which is for "translational coupling", a regulatory mechanism.)

The pipeline maps reads to transcripts.fasta, which is why I think allowing multimapping in bowtie2 will make a difference.

Another one, again would be great if you could review! https://github.com/nf-core/rnaseq/pull/1806