RNA-seq remove any reads that map more than 2 times.

[fabrice]

I am working on RNA-seq Illumina Pair-end data.

Does anyone have a good suggestion that is it neccessary to filter the reads to remove any reads that map more than 2 times?

Thank you.

[Richard Finney]

It's not necessary, beware if both map to multiple places, because RPKM or snp calls there are doubtful. Maybe knowing they're non-specific mappings is good enough. You will not want to spend much time checking out gene fusions between GENEFAMILYMEMBER1 and GENEFAMILTYMEMBER2.

[fabrice]

If both map to multiple places, should we remove them?
I just think if we take these reads into account, we will get highly estimate for some genes.

I am not for snp calls.

It's not necessary, beware if both map to multiple places, and RPKM or snp calls are therefore dubious. Maybe knowing they're non-specific mappings is good enough.

[eslondon]

Multi-mapping reads are quite common when using 35bp short reads. Using paired-end 76bp or 100bp reads it is much less common, because unique assignment is quite precise.

Moreover, I think there was a bit of confusion in the thread. A read can map FULLY in two different places, because the mapping is AMBIGUOUS. In this case it has been shown that one can use a weighted score approach (giving a 0.5 score to each mapping) improving overall expression scores, and similar other strategies.

One can then use several strategies, e.g.:

http://bioinformatics.oxfordjournals.org/content/25/19/2615.short

One recent paper:

http://genomebiology.com/2011/12/2/R13/

It is completely different on the other hand, if a read maps UNIQUELY in TWO FRAGMENTS in TWO LOCATIONS OF THE GENOME. That would be indicative of a gene fusion event, of interest for example in tumor samples.

http://genomebiology.com/2009/10/10/R115/

Elia