Tophat's accepted_hits.sam reports all the ISIZEs equal zero

[ruping]

Hi, I run tophat for pair-end reads, but the results report all the ISIZE equal 0, which is quite confusing:


read1 99 chr1 26773653 255 36M = 26773726 0 GGGATCCCACCCTGGGGACCCCCACGATTGGCCACC ;<A555?<;?6C@D-5@...
read2 147 chr1 26773726 255 36M = 26773653 0 TCACCTCTCTCAGAGCATCTGGCTGTTTAGCAGAAC 3C0@C=C@?-C=?A?C@...
read3 137 chr1 27542972 255 36M * 0 0 GCCTATCCCCATTGGTGTGCATATGCTGTGTGCAGG 3,?33807;*;2;;,>B;B))*756...


based on the definition of ISIZE, if the two reads in a pair are mapped to the same reference (indicated as "=" in the 7th column), ISIZE should be calculated, why it's all zero here?

Thanks in advance for help.

[Cole Trapnell]

The reason is that those are supposed to be (as I understand the SAM spec) the implied length for the paired fragment in that record. However, since the fragments came from the transcriptome, it's hard to know just from the genomic coordinates of the alignment how big the fragment would be in the *transcriptome* coordinate system. Imagine that you had a mate pair where one read is on one exon, and the other mate is on the other, but neither read spans the intron in between those exons. There's no way without knowing which transcript that fragment came from of saying how big that fragment is. Rather than make a wild guess, TopHat puts zero there, since TopHat leaves to Cufflinks the job of figuring out which transcript each fragment came from.

[ruping]

The reason is that those are supposed to be (as I understand the SAM spec) the implied length for the paired fragment in that record. However, since the fragments came from the transcriptome, it's hard to know just from the genomic coordinates of the alignment how big the fragment would be in the *transcriptome* coordinate system. Imagine that you had a mate pair where one read is on one exon, and the other mate is on the other, but neither read spans the intron in between those exons. There's no way without knowing which transcript that fragment came from of saying how big that fragment is. Rather than make a wild guess, TopHat puts zero there, since TopHat leaves to Cufflinks the job of figuring out which transcript each fragment came from.

Thanks, Cole. That make sense.

Sometimes people use the output of tophat as the input of some other tools in which the ISIZE is needed. Anyway, the inserted size can be roughly inferred by users, at least for those cases where the paired reads map closely in the ref. genome.

Thanks again, that really helps.