search

muellan / metacache

memory efficient, fast & precise taxnomomic classification system for metagenomic read mapping
GNU General Public License v3.0
57 stars 12 forks source link

GPU version on bioconda #38

Closed fgvieira closed 1 year ago

fgvieira commented 1 year ago

Dear all,

is there a GPU version of metacache on bioconda?

Can we also define MC_KMER_TYPE for GPU? Beacuse from the code it doesn't look like it: https://github.com/muellan/metacache/blob/4c6db912f92b065ac36d1121a0d4849582896fa3/src/config.h#L44-L48

muellan commented 1 year ago

is there a GPU version of metacache on bioconda?

Not as far as I'm aware of.

Can we also define MC_KMER_TYPE for GPU?

No, the GPU version only supports kmers of length 16 (for various space and efficiency reasons).

fgvieira commented 1 year ago

And do you think kmers of length 16 are enough? Kraken2's default value for nucleotides is 35.

muellan commented 1 year ago

Kraken and Kraken2 use mapping strategies that are different from MetaCache's. While Kraken counts Taxon hits based on single kmers, MetaCache uses multiple kmers within a small region ("window") and thus doesn't require long kmers.

If you want to increase precision you can use the -hitmin <#kmers> option (the default is 4 if I remember correctly) which controls how many kmers in a window are needed for a read to be classified.