[danielsbrewer]

Does anyone know whether bowtie supports aligning multiple read lengths?

I am doing small RNA Solexa sequencing and so after the adapter has been removed I end up with variable length reads. With MAQ it appears that you have to run it multiple times for the different lengths, is bowtie the same?

[doxologist]

thanks for the response.

here's the link to the comparison Heng Li did...

http://maq.sourceforge.net/bwa-man.shtml

[Ben Langmead]

I haven't spent much time looking at BWA and SOAP2, so I'm not the expert. I like Bowtie :-), but I know others who use and are happy with BWA. I haven't met anyone with much experience with SOAP2. There is a post earlier in this thread by Heng Li that talks a little about the key differences among the three. I heard that Heng Li also has some benchmarking results that he's made public somewhere...

Thanks,
Ben

[doxologist]

thanks... that's helpful.

any thoughts on bwa and SOAP2?

[Ben Langmead]

Hi Doxologist,

The circumstances under which Bowtie might miss alignments that are "valid" according to its alignment policy are outlined in the manual (see last paragraph of section "Maq-like Policy"). These misses only occur in -n 2 and -n 3 modes, and they can be avoided by increasing the --maxbts parameter (at the cost of some speed). Unless your read data is very low quality, the fraction of reads missed due to the backtracking limit in -n 2 mode is generally very small (<1%).

Note that when you run 'maq' with -n 2 option (the default), it will find some alignments that actually have 3 mismatches in the seed. Bowtie will *not* report alignments with 3 mismatches in the seed unless -n 3 is specified. It's likely that this is the source of the difference that the anecdotal reports are referring to.

Thanks,
Ben

[doxologist]

quick question: I heard anecdotal reports that bowtie misses a large number of matches compare to programs like Maq. have anyone else heard this? in terms of %, does anyone have any numbers?

[Ben Langmead]

Hi Foram,

I would try upping both the seed length (-l) and the error tolerance (-e). Others may have better suggestions, though. If you find parameters you're happy with, please do post them back here since that will help others.

I'm working on paired-end support currently. Expect it in a few weeks or so.

Thanks,
Ben

Does anyone have any recommendations for scoring params when mapping long (76bp Illumina) reads?

Also, my reads are PE -- any chance this will be supported soon?

[foram]

Does anyone have any recommendations for scoring params when mapping long (76bp Illumina) reads?

Also, my reads are PE -- any chance this will be supported soon?

[zee]

Cole,

Thanks for that clarification. I found that some work I was doing with bowtie seemed to produce a lot more results than what I expected, even when I did use the '--best' option for reporting hits. When I looked at how it stacked up against some other aligners, I felt that there was some kind of overestimation when using the mapping quality score to evaluate good quality SNPs, correct alignments.
I would just need to be cautious how I interpret bowtie's results with respect to metric's derived from other aligners.

[Cole Trapnell]

I was using bowtie and I wanted to compare the mapping qualities when converted to .map format. Would these mapping quality scores be comparable to that for MAQ results?

Bowtie's mapping qualities are only a rough approximation of Maq's. Maq computes mapping quality like this:

Q = min {q_2 - q_1 - 4.343log(n_2), 4 + (3-k')(q_bar - 14) - 4.343log(p_1(3-k,28)) }

Where:

q_2 is the quality-weighted Hamming distance of the best hit, q_2 is the quality-weighted Hamming distance of the second best hit, q_bar is the average quality value on the 5' end of the read, and "p1(k,28) is the probability that a perfect hit and a k-mismatch hit coexists given a 28bp sequence which can be estimated during alignment" (from the Maq paper).

Bowtie, as discussed previously in this thread, doesn't guarantee that it will find the best hit, and by default, won't even continue searching for the second best one. So Bowtie can't really compute Mapping quality this way. Instead, our Maq converter (which was derived from Heng Li's ELAND converter) calculates mapping qualities as follows:

Q = (3 - k) * 25 - log(# of other equally good occurances found by Bowtie).

Where k is the number of mismatches in the seed region of the alignment. This is not as nice as Maq's method. However, it works without forcing Bowtie to be used in one of it's slower modes (--all, for example). In our tests so far, Maq's assembler handles qualities computed this way pretty well and produces good SNP calls.

[florian]

hi zee

sorry, if i gave you the wrong impression, but i'm actually not a developer of bowtie. i cannot claim any of this fame -- unfortunately :-D.
as to the question, i'll better leave the answer to ben, as i'm not sure about the answer either.

cheers,

f

[zee]

HI Ben/Florian

I was using bowtie and I wanted to compare the mapping qualities when converted to .map format. Would these mapping quality scores be comparable to that for MAQ results?

[doxologist]

thanks. looking forward to it.

[Ben Langmead]

Does Bowtie work with colorspace data?

Not yet - that's on the TODO list but we're going to tackle paired-end alignment and gapped alignments first.

Thanks,
Ben

[doxologist]

Does Bowtie work with colorspace data?