Bowtie2 --score-min or mismatch setting

[sanjeevksh]

Dear All,

I am trying to figure out how to set a mismatch (--score-min) value in Bowtie2. On the forum I have seen '--score-min' settings like 'L, -0.5,-0.2', 'C,0,0', 'C,-1,0' and the assigned default setting by Bowtie2 is 'L,-0.6,-0.6'.

I wish to know the reasons behind deciding values for '--score-min' option. Bowtie2 manual mentions some details that it is a function of read length etc. but does not divulge into further details on how does one decide which function to choose as both 'L' (Linear) and 'C' (Constant) have been used in the cases mentioned above as well as the rationale behind the other two parameters.

My dataset is from Illumina Hi-Seq 2500 and reads are 150 bp long. I want to try a range of mismatch settings from 1 to 11 or so with '--very-sensitive' option under the '--end-to-end' mode.

I would highly appreciate any feedback on this.

Regards,
Sanjeevksh

[mastal]

I don't think you can use the --very-sensitive setting if you plan to vary the number of mismatches, I think --very-sensitive corresponds to a certain predefined set of parameters, namely ' -D 20 -R 3 -N 0 -L 20 -i S,1,0.50',
where N is the number of mismatches in the seed region.

[dpryan]

The presets don't alter the score-min setting, so you can mix them at will. The default --score-min will allow up to 15 mismatches for your(I'm ignoring indels here and assuming that you're either using --ignore-quals or that the mismatches occur at positions with high phred scores), so you can gauge the effect by just filtering by MAPQ, the AS tag, or the MD tag.

A constant (C) score-min is only useful if you want to have a maximum edit-distance regardless of read length (again, I'm ignoring the effect of phred scores on the mismatch penalty). In general with end-to-end alignments, you would want the minimum alignment score to change a bit with length. For actually determining the exact equation to use, the main way to do this is with simulated reads, since the --score-min setting directly affects the MAPQ calculation. The defaults generally produce pretty similar MAPQ scores as other programs (namely BWA).

[sanjeevksh]

Hi mastal and dpryan,

Thanks for your replies, are very helpful indeed.

@mastal:
I was not referring to the mismatches in the seed region which -N (0 or 1) actually pertains to but for the entire end-to-end alignment. On the same subject, if say I have to use '--very-sensitive' and only change '-N' parameter, do I submit '--very-sensitive -N 1' or '--very-sensitive' becomes obselete if any parameter covered by this option is altered and one has to submit the entire parameter set separately like '-D 20 -R 3 -N 1 -L 20 -i S,1,0.50'?


@dpryan:
So this means just map once with a default '--score-min' setting and then apply filters and use as it deems fit? Now would try to find out how to filter by MAPQ, AS tag, or MD tag.

I am not specifying anything like --ignore-quals.

Regarding the function to use, I would stick to 'L' which I have used in some test runs.

Regards,
Sanjeevksh

[dpryan]

Yup, exactly. You'll want to filter according to the MD tag, possibly also looking at the sequence in the case of Ns, which you presumably don't want to count.

[sanjeevksh]

MD gives a string so 'XM:i:<N>' field may be more helpful in this context?

Cheers!

[dpryan]

Indeed, yes, I'd forgotten about that tag!