PSGInfer: question for bioinformatics/statistics folks

[rnastar]

Dear all,

I was hoping that one of you may be able to shed some light on a recent publication examining alternative splicing. The publication is entitled "Inference of alternative splicing from RNA-Seq data with probabilistic splice graphs" and is available in Bioinformatics at

http://bioinformatics.oxfordjournals.org/content/29/18/2300.full

The method seems quite novel and seems to provide significant speed improvements over alternative approaches for relative isoform abundances and testing for differential processing between two conditions. This method uses information from all reads in a sample to compute relative isoform levels.

However, I am puzzled at how the method distinguishes between junction spanning reads and non junction spanning reads (reads contained completely in exons). The former provides the only hard evidence of linking between exons, however the notation for their model does not seems to distinguish between junction spanning reads and non junction spanning reads.

In other words, would the method perform similarly if all the junction spanning reads were removed? If so, how does the method utilize junction spanning reads in their model setup?

Thank you!