Does it make sense to limit insert size when mapping on transcriptome?

Question

[Matteo Schiavinato]

As the title says, does it make sense to limit insert size when mapping on transcriptome?

I am mapping RNASeq reads on to a transcriptome and I don't have access to the experimentally designed insert size so I'm estimating it by mapping a subset and observing the mapping distance distribution, to then use it as a parameter in mapping (-I -X in bowtie, for example). However, I don't know if what I'm doing actually makes sense or not. Two out of four of my samples have a very precise and bell-shaped insert size distribution, while one has a little hunch on the small size end and the other is a very broad bell (like from 100 to 500 nt size).

Somebody can shed some light on this?