Tophat Multihit Reads

[allerzbul]

Hi, I'm running the top hat aligner for a ribo-depleted human RNAseq dataset. I'm interested in looking at expression in snRNAs. These loci in particular are very repetitive and when I run Tophat using the default settings (allowing up to 20 hits for each read and reporting these secondary alignments), none of these loci are assigned any single mapping reads. I originally wanted to assign the other multi mappers using the single read count ratios (there is an option for this in Cufflinks) but since there are no single mappers, this probably won't work. I tried an alternative approach, setting Tophat -g 1, which based on my understanding assigns each read to it's best mapping location (based on the mapping score) or randomly, if all of the map scores are the same. This seems to have worked for my snRNAs. For the loci i've looked at, it seems that their ratios when compared to each other are about the same as when I allowed multi-mapping.

I want to know if using the -g 1 has any downfalls? It seems to work really well (based on my initial look) and it removes a lot of the complications that come with dealing with multi mapping reads in differential expression analyses. But if it does work so well, why is default to allow 20 multi mappers? Also, has anyone else come up with different ways to deal with multi mapping reads in differential expression analyses?

Thanks so much for your help!

[swaraj]

This issue is discussed in a publication (http://www.biomedcentral.com/1471-2164/12/516), which reports that accepting multi mapping reads (up to 10) is beneficial (Fig 1). It accounts for maximum
number of gene detection especially genes with low expression levels. The authors argue
that multihits allows the inclusion of homologous sites from the RNAseq data. Using a
reference GTF file with predicted or known transcript models can alleviate the problem. But it is not an ideal case for studies where transcription of repeats is the focal point.
Secondly this combination increases the false negative transcript population
drastically.