Uniquely mapped reads and difference for single end and paired end reads

[gene_x]

I know there have been a few posts on seqanswers and biostars that asked about how to get uniquely mapped reads.

As Heng Li put it,

"Uniquely mappable read" is never satisfactorily defined.

and

Looking at mapping quality is not only the simplest but also the best solution in most cases.

So I thought I would just use a mapping quality cutoff, say 30, to filter out reads to get uniquely mapped reads.

This is simple for single end reads. I'm not sure how to do it for paired end reads since you have two mapping quality numbers. How should it be done?

[dpryan]

There's no general answer to this, it depends on the aligner. Many aligners will give both reads in a pair the same MAPQ as long as they aren't aligned as singletons. In that case, the same MAPQ will work (though a cutoff of 30 seems extreme). Otherwise, it depends on whether the aligner adds an aux tag to one of the alignments. Note that it's completely possible for only one read of a pair to be multi-mapping, such as when one end of a fragment overlaps a simple tandem repeat...though in practice I've not run into this enough to worry about (after all, you have biological replicates, so a bit of noise in each sample should be fine).

[gene_x]

There's no general answer to this, it depends on the aligner. Many aligners will give both reads in a pair the same MAPQ as long as they aren't aligned as singletons. In that case, the same MAPQ will work (though a cutoff of 30 seems extreme). Otherwise, it depends on whether the aligner adds an aux tag to one of the alignments. Note that it's completely possible for only one read of a pair to be multi-mapping, such as when one end of a fragment overlaps a simple tandem repeat...though in practice I've not run into this enough to worry about (after all, you have biological replicates, so a bit of noise in each sample should be fine).

I see. I tried to count the number of properly paired reads with MAPQ either larger than 10 or smaller than 10. I observed about 1% of the read pairs have one read with MAPQ >=10 and the other read with MAPQ <10.

PS: The alignment was done with novoalign.