how to use tophat to map orphan reads from paired-end sequencing

[capricy]

I have a hard-to-obtain sample sequenced. After removing the low quality reads, I found many orphan reads from each end. That is, the read in seq1 is good, but the corresponding read in seq2 has to be removed, vice versa.

I don't want to lose the information of these reads, so plan to map them to the reference using the tophat. I am seeking the best plan to go:

1. separately map them:
tophat genome_index_base seq1
tophat genome_index_base seq2

or

2. can I still treat them as paired end:
tophat genome_index_base seq1 seq2

please note that seq1 and seq2 each contains the orphan reads. But they were originally from the paired-end sequencing...

Thank you very much for any advice!

[colindaven]

I would do both and check the alignment statistics.

I've found STAR to be far quicker than Tophat and give great results in very large genomes, perhaps you can try that as a comparison. It also to my knowledge treats the reads differently, combining R1 and R2 into a 200bp fragment and mapping that. Perhaps that would produce better results with your reads ?

[Brian Bushnell]

If you quality-trim using BBDuk, you can set it to trim reads down to a minimum of 1bp, so that there won't be any orphans:

bbduk.sh in1=r1.fq in2=r2.fq out1=trimmed1.fq out2=trimmed2.fq qtrim=rl trimq=10 rieb=f minlength=20

"reib" means "RemoveIfEitherBad"; when true, pairs are discarded if either fails the filter criteria (in this case, being shorter than 20bp after trimming). When false, pairs are discarded only if BOTH fail.

That way the files can still be mapped as paired. The reads trimmed down to 1bp won't get mapped. But since the sequence is hard to obtain, you could also just skip quality-trimming entirely and try mapping with a program tolerant of errors (like BBMap). Then if some reads are unmapped, trim those reads only and try mapping them again.