Deprecating SSW in scikit-bio: recommendation for alignment algorithms in Python

[ebolyen]

Hello everyone,

We're planning on removing the SSW module from scikit-bio. This will occur sometime before the next release (which should be sometime in October).

It has been a general source of maintenance issues and limits our ability to target Windows platforms. While there have been some very lovely contributions such as #1709 and #1797 (from @mr-c and @theAeon respectively), we think the best approach is to remove this module.

Instead, we recommend using @jeffdaily's library Parasail and corresponding Python bindings: jeffdaily/parasail-python.

The documentation for this package is quite complete, and supports a similar usage pattern as our SSW bindings:
https://github.com/jeffdaily/parasail-python#ssw-library-emulation

The only difficulty you may encounter from this transition is converting the cigar strings (from the "traceable" algorithms, into scikit-bio alignment objects). We would be happy to accept pull requests to this end, but we don't have anything in particular planned.

If you find @jeffdaily's Parasail useful, please remember to cite it :)

[ebolyen]

And here is a worked example comparing skbio's SSW with Parasail:
https://gist.github.com/ebolyen/f759644a67d97a808f0ddb0933bcb4a0

[theAeon]

Would this skbio-ified parasail package (that I shamelessly copied and packaged from @jvkersch) be of any use?

[ebolyen]

That is super lovely @theAeon! Thanks for sharing!

[ebolyen]

I will mention we are looking for maintainers for this project, so if anyone feels particularly strongly about alignment and how that should work in skbio (or other parts of scikit-bio), feel free to get in touch!

[jeffdaily]

Wow, I'm flattered you all still find parasail useful! I wrote parasail as part of my dissertation and always hoped it would help the community. Unfortunately, in my current role I barely have time to maintain what is there. Perhaps someone from this community could help? I don't know how such transitions work; this was my first solo open source project.

[gregcaporaso]

pywfa was pointed out in #254 as another performant Python-accessible aligner.

[gregcaporaso]

We've decided to delay this removal, most likely for one more release. We identified two tools that are using local_pairwise_align_ssw: IAB and q2-feature-classifier. We suspect therefore that there may be others, so we're going to save this for 0.6.0.

[theAeon]

Also makes sense from a semver perspective for sure.