Bowtie maxbts

[Bio.X2Y]

Hi,

I understand that Bowtie offers the 'maxbts' and 'pairtries' parameters to control the trade-off between processing time and comprehensive search.

The following scenario is confusing me: if I map both ends of an ambiguous pair to hg19 independently, setting maxbts to 10billion, I get back 12k hits for end #1 and 14k hits for end #2.

However, if I map the pair as a pair, using 'tryhard' to give a high 'pairtries', I end up getting back pairings where one of the ends maps to a location that wasn't picked up in the original independent search.

My question is: what would cause the independent search to miss matches that the paired search picks up? I presumed that memory and maxbts are the only limiting factors?

Thanks,
Bio.X2Y

Inputs:

File: dodgy_input_1
@HWI-EAS283:2:1:810:1386#0/1
GGGAGTTCGAGACCAGCCTGACCAACATGGAGAAACCCTG
+HWI-EAS283:2:1:810:1386#0/1
]bZ``aaabbbbb`bX`_babbaa`abb`aaaaaa``aaa

File: dodgy_input_2
@HWI-EAS283:2:1:810:1386#0/2
CCCGGGTTCAACCAATTCTCCTGCCTCAGCCTCCTGAGTA
+HWI-EAS283:2:1:810:1386#0/2
a_a_Zbababaa`bbbbababb`babbb]_aaaabbabab

Independent Search:

time /Volumes/Thymine/bowtie-0.12.5/bowtie \
-t \
-n 2 \
-m 100000 \
-a \
-p 4 \
--maxbts 1000000000 \
--best \
--solexa1.3-quals \
--chunkmbs 128 \
hg19 \
dodgy_input_1,dodgy_input_2 \
--un dodgy_pair_independent_unmatched.bowtie \
--max dodgy_pair_independent_ambiguous.bowtie \
dodgy_pair_independent_matched.bowtie \
&> dodgy_pair_independent_console.out

Paired Search:

time /Volumes/Thymine/bowtie-0.12.5/bowtie \
-t \
-n 2 \
-m 1000 \
-a \
-p 4 \
--best \
--tryhard \
--solexa1.3-quals \
--chunkmbs 128 \
hg19 \
-1 dodgy_input_1 \
-2 dodgy_input_2 \
--un dodgy_pair_pe_unmatched.bowtie \
--max dodgy_pair_pe_ambiguous.bowtie \
dodgy_pair_pe_matched.bowtie \
&> dodgy_pair_pe_console.out

Results:

For the independent search, I get 12,637 matches to end #1 and 13,946 matches to end #2.

For the paired search, I get 919 pairs (although I expected 393 since that many pair combinations from the original results can be combined to form pairs that satisfy -I and -X).

Example of result pair that has a mapping for end #2 that did not appear in the independent search:
HWI-EAS283:2:1:810:1386#0/1 + chr11 107533126 GGGAGTTCGAGACCAGCCTGACCAACATGGAGAAACCCTG >C;AABBBCCCCCAC9A@CBCCBBABCCABBBBBBAABBB 344
HWI-EAS283:2:1:810:1386#0/2 - chr11 107533223 TACTCAGGAGGCTGAGGCAGGAGAATTGGTTGAACCCGGG CBCBCCBBBB@>CCCBCACCBCBCCCCABBCBCBC;@B@B 344 11:C>G,13:C>T

[Bio.X2Y]

Hi,

Sorry for bumping this, but I'd still be very interested if anyone has any thoughts on this?

To summarise, I'm finding that if I use Bowtie to map both ends of my paired-end reads independently, it fails to find some matches that it does find when I map them as a pair.

I can't find anything in the manual that would explain this behaviour, and I'd like to get a feel for why it's happening,

Thanks for your time,
Bio

[Ben Langmead]

Bio,

I'm a little confused by the question, but I think the answer is that when you have a pair of mates and one mate lies in a repeat (repeated more times than your -m threshold, say) and the other lies in unique sequence, Bowtie can resolve the "true" location of the repetitive mate. In that way, a mate whose alignments would otherwise be suppressed by -m in unpaired mode might be reported in paired mode.

Hope that helps,
Ben

[Bio.X2Y]

Hi Ben,

Thanks for the reply, but I think what I'm seeing is something slightly different.

To summarise the above, I independently mapped both ends of a single pair using "-m 100000", and got 12,637 matches to end #1 and 13,946 matches to end #2. Since I used tryhard, it implies that neither read has more than 100,000 matches (because if it did, it would be suppressed by -m). It also implies that I should be seeing all the matches.

When I mapped them as a pair, therefore, I didn't expect to see any hits for either end that did not appear in the original search.

Does this mean that even when I use "-m" and "tryhard", I might still be missing out on a lot of matches?

Thanks again,
Bio

[Ben Langmead]

When I mapped them as a pair, therefore, I didn't expect to see any hits for either end that did not appear in the original search.

This is the part I would dispute. You do expect to "recover" some alignments that were suppressed in the original search.

Hope that helps,
Ben