Salmon output: transcripts quantified with zero reads support

Question

I quantified some samples using Salmon. According to the documentation of output format, the last column of the 'Quantification File' represents the number of reads that supported the given transcript, namely:

NumReads — This is salmon’s estimate of the number of reads mapping to each transcript that was quantified. It is an “estimate” insofar as it is the expected number of reads that have originated from each transcript given the structure of the uniquely mapping and multi-mapping reads and the relative abundance estimates for each transcript.

In some case, I had in the output some transcripts that had NumReads=0.000000. Does this mean that I should discard them from the list of transcripts obtained in that sample? How comes that those transcripts end up in the output file in first place, since they don't have any reads support and the quantification was done using the reads themselves?

Answer 1

All transcripts should always be returned in the output, whether there's evidence supporting their expression or not. This is for the sake of simple convenience, since typically one would be merging these values across samples into a matrix. Missing values will then tend to become either NA or cause an error, when what would be appropriate would be to have values of 0. This also facilitates calculations such as "what's the median number of reads per transcript?" or "what percent of transcripts have an expected reads count of at least X".