Pipeline Integration¶

gmlst ships a container image and ready-to-adapt workflow snippets so it drops into large-scale surveillance pipelines (Nextflow, Snakemake, or plain Docker) without local conda setup.

Docker¶

The image (indexofire/gmlst on Docker Hub) bundles gmlst plus all four alignment backends (BLAST+, minimap2, MUMmer4, KMA), KMC, pyrodigal dependencies, and samtools:

# Pull the release image
docker pull indexofire/gmlst:latest

# Type one sample (mount the working directory; schemes cache in /home/mambauser/.cache/gmlst)
docker run --rm -v "$PWD:/data" indexofire/gmlst:latest \
    typing mlst -s saureus_1 /data/sample.fna -o /data/result.tsv

# Batch with 16 parallel samples and reuse a persistent scheme cache
docker run --rm -v "$PWD:/data" -v "$PWD/.gmlst-cache:/home/mambauser/.cache/gmlst" \
    indexofire/gmlst:latest \
    typing mlst -s saureus_1 -t 1 --max-workers 16 /data/assemblies/*.fna -o /data/batch.tsv

Version-pinned images are tagged per release (indexofire/gmlst:0.5.2, 0.5, …). The image build is smoke-tested (gmlst --version, blastn, minimap2, kma) before push — see .github/workflows/docker.yml.

Build locally from a checkout:

docker build -t gmlst:local .
# Or pin an exact PyPI version:
docker build --build-arg GMLST_VERSION=0.5.2 -t gmlst:0.5.2 .

Nextflow¶

See examples/pipelines/gmlst_mlst.nf:

nextflow run gmlst_mlst.nf \
    --samples "assemblies/*.fna" \
    --scheme saureus_1 \
    -profile docker \
    -resume

The process emits per-sample TSV + JSON and a versions.yml for provenance. Pass extra CLI flags via --args "--minscore 50". On cloud executors let Nextflow fan out per sample; inside each task gmlst stays single-sample (no --max-workers needed).

Snakemake¶

See examples/pipelines/Snakefile:

snakemake -j 16 --use-singularity --config scheme=saureus_1 sample_dir=data/assemblies

Parallelism guidance¶

--max-workers N is the main speed lever for single-host batch runs (see the measured table in the README).
In workflow engines, prefer task-level fan-out (one sample per task) — each task then runs gmlst with a single sample and default threading, which keeps resource requests predictable.
Mount a persistent cache (GMLST_CACHE_DIR or the default ~/.cache/gmlst) so schemes download once per pipeline instance, not once per task.

Environment variables in containers¶

Variable	Typical container value
`GMLST_CACHE_DIR`	a mounted volume, e.g. `/data/.gmlst-cache`
`GMLST_TMPDIR`	scratch space on a mounted disk for large batches
`GMLST_PUBMLST_API_KEY`	secret injection for post-2024 PubMLST data

See the Configuration Reference for the full variable list.