Tophat: uniquely mapped or multi-hit?

[mrfox]

hi all,
I wonder how many hits you guys allow when using Tophat to map RNA-Seq reads. For me, I always use uniquely mapped reads(-g is set to be 1). However, I am not sure if this seems to be too conservative and thus results in potential true signals. The dilema is, if multi-hits are allowed, what is a good cutoff for the maximum hits? And for the downstream analysis(gene expression, splicing), not all the software tools can well handle multi-hit correction.

I would like to listen to your comments and advice.

[swaraj]

If you have a reference trasncriptome to use while mapping with Tophat, then I suggest to leave -g to its default value and use --prefilter-multihits. This takes care of many of the multiple mappings. Further at the cost of speed the --read-realign-edit-dist can be set to zero to ensure mapping of each read at each alignment step. This should ensure a more accurate mapping.

For further downstream filtering Cufflinks can be run with --max-multiread-fraction 0.5 instead of 0.75 to ensure no transcript with more than 50% multiread mappings is built. But I think that is not a necessary step.

Actually a -g of 10 can be good to maximum number of genes especially lowly expressed ones as reported by this publication

http://www.biomedcentral.com/1471-2164/12/516