Optimum insert size for De-novo transcriptome assembly

[tellsparck]

I am making RNA-Seq libraries from mouse tissue for de-novo transcriptome assembly. I am planning to use cufflinks for this (the genome guided denovo assembly approach). What insert size is optimal for this? Since we are aligning to a well annotated genome, does it make sense to have a higher average insert size? This way, many more reads will span more than one exon with paired end sequencing, Won't this boost chances of finding alternative spliced isoforms? But does this entail sacrificing info on 3' and 5' end which is equally important (an important goal of our study is to find alternate 3' ends of genes)?
Thanks for your inputs!

[Brian Bushnell]

I think a broad range of insert sizes is good, particularly if you are interested in short transcripts that may not be represented at all if you filter out short inserts. The less size-selection, the less bias (aside from remove fragments that are too short to be useful or too long to amplify), and the greater the library uniqueness, which is important for assembly.