Force insertions to align with codon boundaries

[donkirkby]

In the 09-Dec-2016.M01841 run, sample PNL4-3-1-V3LOOP_S73 reported a strange consensus sequence.

After digging into the problem, we found that a six-base insertion wasn't aligned with the codon boundaries. Here's a portion of the HIV1B-env reference we use, with the usual insertion marked with dashes:

AGTATACATATA------GGACCAGGG

The amino acid equivalent is:

SIHIGPG

Here's a portion of one of the sample's reads, as bowtie2 aligned it with the reference below:

AGTATCCGTATCCAGAGGGGACCAGGG
AGTATACATAT------AGGACCAGGG

You can see that the insertion is base to the left of its usual position. That pulls the G from the right side of the insertion over to the left, and is translated as the following amino acid sequence:

SIRMGPG

One possible solution: if an insertion's length is a multiple of three, then force it to align with the codon boundaries. For now, just choose the closest codon boundary.

[donkirkby]

Possible solutions discussed at development meeting: combine identical reads in sam2aln step, then look for deletions that don't align to codon boundaries. When you find them, try using Smith-Waterman alignment, possibly with a penalty for missing the codon boundary. There is a Python library called swalign that might be helpful.

[greenbeard1]

I saw this paper and thought i'd pass it along in case it's useful here: http://bib.oxfordjournals.org/content/early/2010/05/11/bib.bbq015.full?view=long&pmid=20460430

…

On Thu, Dec 15, 2016 at 11:29 AM, Don Kirkby ***@***.***> wrote: Possible solutions discussed at development meeting: combine identical reads in sam2aln step, then look for deletions that don't align to codon boundaries. When you find them, try using Smith-Waterman alignment <https://en.wikipedia.org/wiki/Smith%E2%80%93Waterman_algorithm>, possibly with a penalty for missing the codon boundary. There is a Python library called swalign <https://pypi.python.org/pypi/swalign/> that might be helpful. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#379 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AFl7ZFyoi77LLLTK5zW5_XYIUW2DexbUks5rIZUmgaJpZM4LLLBL> .

[donkirkby]

Might also want to look at the strange result described in issue #331 when considering options for this issue. The Gotoh algorithm described in that issue is currently used to align the consensus sequence to the coordinate reference.