Reads not mapped in proper pair: Bowtie output

[RGP]

Hello,

when I align paired end fastq reads mapping to two different chromosomes using BWA, I typically get this result:

ILLUMINA-8B7A89:1:33:5159:13399#TAGCTTA#0 65 chr22 21854387 37 30M chr9 132720172 0 TGGCCCAACGATGGCGAGGGCGCCTTCCAT
GEFGGGEBEEEEEEEEEAEECDC==BCCBA

ILLUMINA-8B7A89:1:33:5159:13399#TAGCTTA#0 129 chr9 132720172 37 30M chr22 21854387 0 TCTGCTGAGCAGCGGGATCAATGGCAGCTT
FBFDEAEGEGGGEGGGDF5GGGGGGGGBGD


where BWA correctly reports the alignment and labels the read as 'not mapped in proper pair' (flag 2 off).

When I try the same using Bowtie I get a completely different result, with the paired read flagged as unmapped:

ILLUMINA-8B7A89:1:33:5159:13399#TAGCTTA#0 77 * 0 0 * * 0 0 TGGCCCAACGATGGCGAGGGCGCCTTCCAT GEFGGGEBEEEEEEEEEAEECDC==BCCBA

ILLUMINA-8B7A89:1:33:5159:13399#TAGCTTA#0 141 * 0 0 * * 0 0 TCTGCTGAGCAGCGGGATCAATGGCAGCTT
FBFDEAEGEGGGEGGGDF5GGGGGGGGBGD


Notably, when I map the two fastq as single reads, Bowtie reports the correct mapping to Chr9 and 22. I tried to play with Bowtie parameters with no luck. Of course I could try to output the 'unmapped' reads and remap as single reads but this is more complex and highly inefficient. Is there any Bowtie setting that allows me to correctly map 'flag-2-off' reads?


Thank you in advance,
Rocco

[gringer]

Notably, when I map the two fastq as single reads, Bowtie reports the correct mapping to Chr9 and 22

This doesn't sound at all like a correct mapping, unless you have a chromosome-level genome modification. It doesn't surprise me that bowtie flags these as unmapped, because it has a limit on how far apart it will search for the matching pair.

[RGP]

Hello Gringer,

thanks for your reply. What I'd like to highlight here is the difference between 'Unmapped' (i.e. SAM flag 0x0004 and 0x0008) and 'Not mapped in proper pair' (0x0002). The two sequences are (or should be) mapped, thus the unmapped flags should be set off and the corresponding Chrs and position reported in the SAM output.
As you correctly stated, the 'distance' between the two reads is definitely abnormal, so I would expect a 0x0002 flag off.
If you look at the BWA output, it reports the paired read with a 65/129 flag, where:

65:
Read paired
Read not mapped in proper pair ***
Read mapped ***
Mate mapped ***
1st in pair

129
Read paired
Read not mapped in proper pair ***
Read mapped ***
Mate mapped ***
2nd in pair

Please, note that the two sequences ARE mapped although they are not in proper pair.
This sounds more logical to me. If you know a way to force Bowtie to report like this, please let me know!


Regards,
Rocco

[gringer]

well, I'm not sure of any workarounds in Bowtie, but I agree that this is an incorrect output and I've submitted a bug report:

https://sourceforge.net/tracker/?func=detail&aid=3364233&group_id=236897&atid=1101606

[gringer]

I've just had a response for the bug report:

Bowtie 1 is not smart enough to notice or report when mates from a
paired-end read align non-concordantly. In Bowtie 1, either a pair aligns
concordantly or it fails to align. Thus, I don't think the issue you raise
comes up. Bowtie 2, however, is smart enough to try to align the mates
separately, and it reports the correct SAM flags in that situation.