TopHat2 doesn't map spliced reads

[bioserv]

Hi,
I am using TopHat2, version 2.0.6, and many of the reads in my data that should be aligned to two exons (with a gap between them) are either mapped sequentially to the genome or unmapped. I tried giving TopHat a specific artificial sequence that spans two exons (with a GT-AG intron in between) in multiple copies that vary by one nucleotide at the ends, but with the same junction (and most of the exons) sequence - TopHat managed to align one of these copies and the rest were unmapped.
Can anyone advise?
The TopHat parameters were -a 5, --segment-length 18 and the artificial sequence is GCTCTCTTGTGTAAATGAACCACATTTTCTGTATCCTTTATTTGGGTTTGCGATCACAGTAATGGAGGAGTACTCCCTTGCTCTACATCCTCACCATCATGTGCTGT
(maps to mouse chr6:67,557,197 in mm10).
Thanks, Yuval.

[dpryan]

Did you give tophat an annotation file? That will normally increase the rate of spliced-read alignment. See also the graphs in this paper. In particular, note the jump in accuracy when given an annotation file in figure 3:

[bioserv]

Hello dpryan,
Thanks for your suggestion, adding a GFF indeed makes all copies accepted hits. However, it is still strange that without the GFF one copy is aligned while 9 others are unmapped...

[dpryan]

I'd have to look into the underlying sequence in that area to see exactly why this occurring. In general, if the read can align with a valid alignment score without being spliced, then tophat2 won't resort to splitting the read into segments prior to mapping unless you provide it with an annotation (in which case, it aligns to the transcriptome first, though removing that gene from the annotation would produce equivalent results).