question about the sam format

[capricy]

Hello, there,

I used tophat to map the reads to the genome reference and the got the aligned bam file. I have a question about the following reads:

====
grep "HWI-ST514:143982632:C37PRACXX:7:1101:10044:92913" accepted_hits.bam.sam
HWI-ST514:143982632:C37PRACXX:7:1101:10044:92913 97 A_Cont6 926942 50 61M102N4M A_Cont42 348362 0 TCACATGCTCGACCCTGCTCATGCTTCGCTCATCGAAAAGGCTCAGTTCTTCTACATCGCTGGAT IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII AS:i:0 XM:i:0 XO:i:0 XG:i:0 MD:Z:65 NM:i:0 XS:A:+ NH:i:1
HWI-ST514:143982632:C37PRACXX:7:1101:10044:92913 153 A_Cont6 926970 50 33M102N32M * 0 0 CTCATCGAAAAGGCTCAGTTCTTCTACATCGCTGGATTCTTCTTGACTGTCTCGCCTCCATCAAT IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII AS:i:0 XM:i:0 XO:i:0 XG:i:0 MD:Z:65 NM:i:0 XS:A:+ NH:i:1
====

My understanding is that these two reads are paired. For the first read, according to the flag value and field 7 (A_Cont42), its mate should be mapped, indicating the both reads are mapped. However, for the second read, flag value and field 7(*) indicate that its mate is not mapped.

I feel there is discrepancy here. Could anyone help me figure out why?

Thanks a lot!

[dpryan]

That appears to be a bug in tophat. You're absolutely correct that that pair of alignments makes no sense. Given how tophat works, I suspect that it's incorrectly merging together alignments from one of the stages in the alignment process (namely from the split and map the mates separately across splice junctions step).

[capricy]

looks like we can not trust the information about the mates in tophat output...

[dpryan]

looks like we can not trust the information about the mates in tophat output...

At least that version of tophat. If you're using the most recent version and issue a bug report then perhaps this can get fixed for the next release.

[Mchicken]

That is a common problem of TopHat2 that I also observed two months ago. It is also a problem for some of the piccard-tools which you might run on your SAM files.

Although the bitflag of paired end runs is misleading for some information, you can still infer the strand information (+/-) and the position in pair ("first in pair" / "second in pair"). If you still want to know if a pair of reads is proper mapped, simply verify that the two reads are mapped on opposite strands on the same chromosome and check if the distance between the mapping positions is within the range of your expected insert size. That is what I did in my analysis.