MRG: update FAQ answer on k-mer size

[ctb]

A collaborator asked why I was using k=21 for a metagenome gather instead of k=51, and I realized that the FAQ answer on k-mer sizes was too simple.

This updates the FAQ answer to the following:

What k-mer size(s) should I use with sourmash?

The short answer is: for DNA, use k=31.

Slightly longer answer: when we look at the k-mer distribution
across all of the bacterial genomes in GTDB, we find that 99% (or
more) of 31-mers are genome, species, or genus specific.

If you go lower (say, k=21), then you get a few percent of k-mers
that match above the genus level - family or above. This is useful for
fuzzier matching, e.g. if you have a metagenome with a high fraction of
unknown k-mers.

If you go higher (k=51), a higher percentage of k-mers are genome-specific.
This can be valuable in situations where you have highly specific reference
genomes (e.g. isolates, single-cell genomes, or MAGs) that should match
very closely to this metagenome.

For the core sourmash operations - search, gather, and compare - we
believe (with evidence!) that (a) the differences between k=21, k=31,
and k=51 are negligible; and that (b) k=31 works fine for most
day-to-day use of sourmash.

We also provide Genbank and GTDB databases for k=21,
k=31, and k=51.

see the PR for the diff ;)

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (4a72236) 86.51% compared to head (99e6496) 86.51%.

Additional details and impacted files

@@           Coverage Diff           @@
##           latest    #2899   +/-   ##
=======================================
  Coverage   86.51%   86.51%           
=======================================
  Files         135      135           
  Lines       15252    15252           
  Branches     2622     2622           
=======================================
  Hits        13195    13195           
  Misses       1758     1758           
  Partials      299      299           

Flag	Coverage Δ
hypothesis-py	25.80% <ø> (ø)
python	92.73% <ø> (ø)
rust	59.00% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ctb]

bump @sourmash-bio/devs

[ccbaumler]

LGTM

[ctb]

thanks!