with --protein, translate whole sequence and then emit k-mers

[olgabot]

This may be a solution for #659

Instead of emitting DNA k-mers and then translating them, which for all but the 0 frame can run into issues of creating an X codon at the end, what about translating the entire read/sequence in 6 frames, and then emitting k-mers? This would have fewer X amino acids output and potentially be more meaningful for translated nucleotide sequence.

[olgabot]

I.e. instead of making DNA kmers and then translating them:

Translate the whole sequence and get k-mers from that:

[ctb]

I implemented this for spacegraphcats, on a trial basis - see spacegraphcats/spacegraphcats#379, script query-unitigs-prot.py.

[ctb]

(and it was quite simple and I liked it :)

[bluegenes]

@luizirber and I talked through the relevant rust code -- it seems we're already translating the entire sequence first and then emitting k-mers.

Here we find the forward 3-frames:
https://github.com/dib-lab/sourmash/blob/eb2b210d40b1441a1467ec507e920339e0bfe437/src/core/src/signature.rs#L96-L105

translate to amino acid
https://github.com/dib-lab/sourmash/blob/eb2b210d40b1441a1467ec507e920339e0bfe437/src/core/src/signature.rs#L106

then kmerize and hash.
https://github.com/dib-lab/sourmash/blob/eb2b210d40b1441a1467ec507e920339e0bfe437/src/core/src/signature.rs#L108-L111

Then, the revcomp 3 frames are done in the same fashion:
https://github.com/dib-lab/sourmash/blob/eb2b210d40b1441a1467ec507e920339e0bfe437/src/core/src/signature.rs#L113-L119

Luiz pointed out that we could improve efficiency by checking for each k-mer (e.g. in a hash table) before hashing (= modify lines 108/116), but otherwise the code is already doing what is suggested here!

[ctb]

closing, since it seems like this is already happening!