Unique VS Non-Unique read analysis

[samt]

I have been debating the it is more appropriate to use only uniquely mapping reads in analysis (transcript variation, expression, structural variation). I wanted to get feedback from other people and poll the consensus for analyzing rna-seq data.

If this has been discussed already, a paper link or thread would be helpful.

[swbarnes2]

There was a recent Nature paper that mentioned uniqueness filtering...some commentary on it here:

http://www.massgenomics.org/2009/09/capture-and-illumina-sequencing-of-human-exomes.html

the blog author's take: you should get rid of non-unique reads, but this is going to limit your max coverage of single end data to 2x the read length. With paired ends, since you can get fragments of multiple sizes, your max coverage can be a whole lot higher.

[samt]

much appreciated, there is some good insight in this paper. if you know of any other papers, link them also

Thanks