cufflinks, tophat, and mysterious snoRNAs

[JueFish]

We have been looking at different tissue transcriptomes and noticed that when we estimate fpkm values (using RefSeq as an annotation) we get lots of snoRNAs as our most highly expressed entities (and by a lot); however, when we put the same data through a pipeline using Tophat, all those different loci are reduced to, usually, one snoRNA locus and that value is brought back into the same expression value range as other highly expressed loci. Anythoughts on why the extreme difference between the two analyses? Given the implementation of Tophat, we would suppose it has something to do with splicing or providing a junction file for mapping, but we're not sure why this should make such a huge difference in our fpkm values.

Thanks.