More loose criteria for bowtie2 mapping

Question

[vasilislenis]

Hello everyone,

I am using bowtie2. I would like to use more loose mapping criteria in order to have more reads that are mapping on a particular place (e.g. mitochondrial genome) of the reference, partially and with more than one mismatches. Is there any other parameter that I can check except the maximum fragment length (-X) that I can give a very large number and the max mismatches in seed alignment (-N) which as I can see it has to be 0 or 1 (so, only 1 mismatch?)

Thank you very much and I'm sorry for my ignorance.

[Ashutosh Pandey]

Have you tried "--very-sensitive" mode?

[vasilislenis]

Very sensitive mode isn't it for more stringent alignments? (Maybe I have misunderstood the parameters). I want the opposite, to have more loose criteria. More reads that are aligned in mitochondria (even partially), so I can have more "material" for a denovo assembly.

[Ashutosh Pandey]

You can try '--local' mode which will allow for partial alignments (soft-clipping of the reads). It should increase the number of aligned reads.

[vasilislenis]

Actually I have tried the --local mode but I was thinking maybe somehow if I could make it more loose. I had -X 1800. Now I think that maybe -X 2400 and -N 1 will make some difference.

Answer 1

[Devon Ryan]

Aside from using local vs. end-to-end alignment, you can also play with the --score-min settings. This will dictate the maximum number of mismatches allowed as a function of mapped length. Note that you also likely want the -a option at that point.

Note that changing around -X to biologically unlikely values is unlikely to benefit you much. While you can play around with the seed size, keep in mind that you'll need to increase -D, though that's at least partially done by -a (no clue how much).

Answer 2

[thackl]

You misunderstood the parameters: very-sensitive == very loose

Your best choice probably is --very-sensitive-local

-N (0) only controls the number of mismatches in the seed, which is given by -L (20). The total amount of possible mismatches/gaps depends on the score given by -S. In --very-sensitive mode, you need a 20 bp exact match and the low score cutoff allows for quite some mismatches/gaps (depending on scoring scheme).

Also, you should not rely on reads that mapped in pairs. Rather try to use every pair with at least one aligned read.

Alternatively, you could use bwa mem, instead of bowtie2. It does local alignments by default and is quite sensitive.

Lastly, if you are aiming for a set comprising an as complete as possible mitochondrium read set, there is another approach: Try to estimate the coverage of your mito in your read set, e.g. from mappings or using a kmer approach. Then downsize your library to hold about 100X mito (e.g. just take the first few Gbp). Assuming your mt is much higher covered than your host genome, this will dilute your host data. Often error-correction and Assembly of this 100X mito already produce quite good mito contigs (and some host repeats, which are usually easy to filter)

Answer 3

[Brian Bushnell]

BBMap is substantially more sensitive than other short-read mappers even at the defaults, so I recommend you try that. You can increase the sensitivity even more with the "slow" flag or by reducing the kmer length.

Answer 4

[predeus]

There's a useful combination described here:

bowtie2 --local -D 20 -R 3 -L 3 -N 1 -p 8 --gbar 1 --mp 3

Worth a try, I think.