How to deal with paired end reads with one mapped, the other not

[shangzhong0619]

Hi all,
I am using gsnap to map paired end reads to reference, and got sam file as output. However, it seems that in the file, there are some pairs with only one end mapped to the reference, the other is not. In this case, should I treat this pair as mapped, or not? Since for paired end reads, we consider pair together, what is the meaning of using pairs with one end mapped, the other not?
Any comments would be appreciated.

[TiborNagy]

This could be a sequencing error, structural variation (where the insert size is differ) or alignment error. If the reference sequence contains ambiguous nucleotides the aligner won't put reads that position (even if the other end can be mapped)

[dpryan]

It generally makes sense to keep those in. Often the second mate of a pair will have completely crap quality, causing it to not align anywhere. This can also happen if you misspecify insert sizes, though unless you get a huge proportion of this then that's an unlikely cause. Then there's unknown SNPs, structural variation, ...

If you're interested in counting genes for differential expression or similar, then go ahead and keep these orphaned reads.

[shangzhong0619]

It generally makes sense to keep those in. Often the second mate of a pair will have completely crap quality, causing it to not align anywhere. This can also happen if you misspecify insert sizes, though unless you get a huge proportion of this then that's an unlikely cause. Then there's unknown SNPs, structural variation, ...

If you're interested in counting genes for differential expression or similar, then go ahead and keep these orphaned reads.

Hi Devon,
what about finding SNPs? Is it useful to include those orphaned reads when finding SNPs?

[dpryan]

Sure, I see no problem with that.