Transcript orientation information to build index

[mriken]

Hi,

I looked into building my own transcriptome+genome index for Salmon. When extracting the fasta sequences for the transcripts, the default behavior is to ignore the strand information, when done via bedtools. All examples I have seen so far seem to ignore the strand; however, shouldn't it be taken into account for a correct 5' -> 3' orientation?

I do see differences in quantification of some transcripts, depending on whether my index is built from strand-aware or strand-unaware fasta sequences. Typically, these are snoRNA transcripts or duplicated/inverted genes and that makes me think it is to do with the directionality of transcripts (although, what I map is PE unstranded reads).

What is the correct way to build my own indexes?
Thank you