Skip to content

Calculating taxonomic purity of extracted bins

This is documentation on how to use scripts/tax-classify.py to estimate the taxonomic purity of bins of k-mers.

Getting set up.

First, we'll need an LCA database, which combines signatures with taxonomic information; the command sourmash lca index builds such a database.

For this we need two things:

  • a bunch of signatures, with names;
  • the taxonomic information for those signatures, calculated e.g. by dib-lab/2018-ncbi-lineages;

For the podar data set, the taxonomic lineages are here: spacegraphcats/data/podar-lineages.csv

To calculate the signatures for the podar data set, we should use a lower scaled value, which will give higher resolution to the taxonomic classification:

sourmash compute -k 31 --scaled=100 {?,??}.fa -f --name-from-first

Then, to calculate an LCA database, do:

sourmash lca index podar-lineage.csv podar-ref.scaled100.lca.json \
    {?,??}.fa.sig \
    --scaled 100 -C 3 --split-identifiers

This will result in a file podar-ref.scaled100.lca.json that you can use with tax-classify.py.

Running tax-classify on catlas region output

First, generate the .node_mh file from a catlas with the appropriate scaled value; it should be no smaller than the one used when computing the signatures & building the LCA database, above.

python -m search.characterize_catlas_regions twofoo_k31_r1 twofoo.vec --scaled=100

Then:

scripts/tax-classify.py twofoo.vec podar-ref.scaled100.lca.json

Running tax-classify-sigs on catlas search output

You can run the same code as in tax-classify.py on signatures produced by search.extract_nodes_by_query, using a different front-end script:

scripts/tax-classify-sigs.py ./podarV_k31_r1_search_oh0_jan19/[012].fa.contigs.sig \
    ~/dev/sourmash/podar-ref.1k.lca.json

Here, the output is more detailed because we expect more of these bins to be low purity.