[Ben Langmead]

Sorry for the confusing question.
If I specify -k 4 --best, multiple alignments will be reported even one best hit is found. I'm wondering is it possible only the best hit (only 1) is report for this case, rather than 4 alignments reported in best-to-worst order.

Please take a look at the documentation for the --strata option in the manual. If that doesn't do what you'd like, please post a small example where Bowtie won't report what you'd like.

Ben

[liu3zhen]

I don't follow your question; please post an example where Bowtie's output is not what you expect/want.

Ben

Sorry for the confusing question.
If I specify -k 4 --best, multiple alignments will be reported even one best hit is found. I'm wondering is it possible only the best hit (only 1) is report for this case, rather than 4 alignments reported in best-to-worst order.

[Ben Langmead]

Thanks, Ben. Bother you again for more questions:
I think Bowtie is more like BLAST in term of output. The specified maximum alignmens were reported in the order of best-to-worst. Does Bowtie have a way to report only the "equal" best hits. If one best hit is found, only report one alignment. If multiple best hits are found, report up to the specified maximum number of alignments. This is important because I know which reads really hit multiple places in the genome by doing so.

I don't follow your question; please post an example where Bowtie's output is not what you expect/want.

Ben

[liu3zhen]

A subset (a subset of 1 in the case of -k 1) is chosen at random.

Thanks, Ben. Bother you again for more questions:
I think Bowtie is more like BLAST in term of output. The specified maximum alignmens were reported in the order of best-to-worst. Does Bowtie have a way to report only the "equal" best hits. If one best hit is found, only report one alignment. If multiple best hits are found, report up to the specified maximum number of alignments. This is important because I know which reads really hit multiple places in the genome by doing so.

[Ben Langmead]

1. Is it possible some reads only have 2 hits even -k 4 and --best are specified?

Yes.

2. How does Bowtie deal with several "equal" best alignments but -k 1 --best were specified? This is the question for how Bowtie deal with repetitive reads.

A subset (a subset of 1 in the case of -k 1) is chosen at random.

Thanks,
Ben

[liu3zhen]

That's right; --best does not limit the number of alignments Bowtie reports. If you ask for 1 alignment (default), --best guarantees it's the best. If you ask for -k 4, --best guarantees they're the 4 best, reported in best-to-worst order. If you ask for -a, --best guarantees that you'll get all of them in best-to-worst order.

Thanks
Ben

Thank for quick reply.
More questions for this topic:
1. Is it possible some reads only have 2 hits even -k 4 and --best are specified?
2. How does Bowtie deal with several "equal" best alignments but -k 1 --best were specified? This is the question for how Bowtie deal with repetitive reads.

Thanks again.

[Ben Langmead]

1) Is it possible to obtain an alignment for a read pair where one read uniquely maps but the other doesn't? (my guess: no)

Definitely yes! That's exactly where paired-end sequencing pays off . If either read aligns uniquely, that alignment will be used as an anchor to look for the mate's alignment and, if it's found, that paired-end alignment will be reported.

2) Does the mismatch setting apply to both reads or are they taken together. In other words if 1 mismatch is specified, can both members of a pair each have 1-mismatch? (my guess: yes)

The mismatch setting applies to each read. So, yes, if -v 1 is specified, *both* mates are allowed to have a mismatch.

Hope that helps,
Ben

[Ben Langmead]

But the manual said that if -k or -a >1 and --best are specified, only best alignments will be reported and they are appear in best-to-worst order, which means that the best alignments are not "equally best".

That's right; --best does not limit the number of alignments Bowtie reports. If you ask for 1 alignment (default), --best guarantees it's the best. If you ask for -k 4, --best guarantees they're the 4 best, reported in best-to-worst order. If you ask for -a, --best guarantees that you'll get all of them in best-to-worst order.

Thanks
Ben

[ecabot]

are pairs considered separately wrt mismatches and uniquness with soap-like policy

I have a couple of questions about how Bowtie deals with mismatches in a paired end run. (Using -v 1 and -m 1). I have my guesses as to how things work, but I am hoping that someone knowlegeable (e.g. Ben) will ring-in with the correct information.

1) Is it possible to obtain an alignment for a read pair where one read uniquely maps but the other doesn't? (my guess: no)

2) Does the mismatch setting apply to both reads or are they taken together. In other words if 1 mismatch is specified, can both members of a pair each have 1-mismatch? (my guess: yes)

[Ben Langmead]

A question for number of mismatches. I can not set up -v 4. (error: -v arg must be at most 3) Does that mean Bowtie at most allow 3 mismatches for whatever length of reads? Thanks.

Hi liu3zhen,

To allow more than 3 mismatches in the alignment, use the Maq-like options: -n/-l/-e instead of -v.

Thanks,
Ben

[Ben Langmead]

The data quality is not optimal but i guess that having no alignment using paired end is not due to that fact and probably parameter should be tuned.
Any one could give me some insight about the optimal setting for the paired end alignment ?
Thanks in advance,
Best,
ramzi

Hi Ramzi,

The options you're looking for are almost certainly -I/-X and --ff/--fr/--rf. You need to have a reasonably good idea of the expected insert size and specify an appropriate range with -I/-X. You should also confirm that your paired-end protocol produces pairs in the fw/rev orientation. This is the typical configuration for Illumina. If your paired-end data has a different orientation, change it with --ff or --rf.

Hope that helps,
Ben

[liu3zhen]

Another question:

I'm reading the manual for -k -a and --best.

I'm confusing about if we put (-k or -a) with --best together. I thought that if a read has several "best" alignments, these "best" should have kinds of "equal" alignment scores. But the manual said that if -k or -a >1 and --best are specified, only best alignments will be reported and they are appear in best-to-worst order, which means that the best alignments are not "equally best".

Hopefully get your help soon, thanks.

[liu3zhen]

A question for number of mismatches. I can not set up -v 4. (error: -v arg must be at most 3) Does that mean Bowtie at most allow 3 mismatches for whatever length of reads? Thanks.

[ramouz87]

Hi,
I'm New in the field of NGS (was working mainly on microarray data analysis) and i'm starting to invastigate comon tools related to sequence analysis.
I have human data (paired reads/ 75 base) and used Bowtie for the alignment.
I used standard parameter for alignment :
bowtie -t -p 8 h_sapiens_37_asm ./s_8_1_sequence.fq ./s_8_1_sequence.fq.bowtie.align
bowtie -t -p 8 h_sapiens_37_asm ./s_8_2_sequence.fq ./s_8_2_sequence.fq.bowtie.align
bowtie -t -p 8 h_sapiens_37_asm -1 ./s_8_1_sequence.fq -2 ./s_8_2_sequence.fq ./s_8_sequence.fq.bowtie.align

and I get respectively the following results:
# reads processed: 6660511
# reads with at least one reported alignment: 4615451 (69.30%)
# reads that failed to align: 2045060 (30.70%)
# reads with at least one reported alignment: 5050548 (75.83%)
# reads that failed to align: 1609963 (24.17%)
# reads with at least one reported alignment: 13371 (0.20%)
# reads that failed to align: 6647140 (99.80%)

The data quality is not optimal but i guess that having no alignment using paired end is not due to that fact and probably parameter should be tuned.
Any one could give me some insight about the optimal setting for the paired end alignment ?
Thanks in advance,
Best,
ramzi

[Ben Langmead]

I am still wondering whether the seed region is defined only for counting the mismatches or not.

Yes. The setting for -l matters for the -n limit but not for the -e limit.

If I want to just use the quality score criterion, and set -l equal to 0, does it work?

No, -l must be set to 5 or greater.

Ben