So if you have a look at this image here

Salmon wrongly is counting this transcript (viral genome) due to the presence of just these poly A reads. There is no other reads which align or map to any other section of the transcript but since this is just a poly A section I think I'd be justified in saying it is counted in error. I'd ideally want a way to set a minimum coverage so that I'd need say 20-50% minimum of a transcript to have some coverage before its accepted as a read.

My script is:

for fn in SRR{1..50};
do
echo "Processing sample ${fn}"
salmon quant -i ../indexes/index -l A \
-1 split/${fn}/${fn}_1.fastq.gz \
-2 split/${fn}/${fn}_2.fastq.gz \
--validateMappings \
--writeMappings=./sams/${fn}_viral  \
 -o quants/${fn}_quant
done

edit (couldnt get hyperlink working)

Wouldn't it be "safer" to trim trailing polyA sequences of a certain length from the reads directly? That way one would probably still get alignments where the polyA is flanked by non polyA reads and the true origin of the read is not a polyA-tail (given read length is sufficient which is seems to be here).