cdm_skani_gtdb

CTS (CDM Task Service) job wrapper for skani paired with the GTDB R232 reference sketch database. Lets you query user genomes against the full GTDB R232 representative set for nearest-reference ANI in seconds (vs hours for the full GTDB-Tk pipeline).

For ad-hoc skani usage without GTDB reference data (user-vs-user comparison, custom reference sets), use cdm_skani.

Container

• Published to ghcr.io/kbaseincubator/cdm_skani_gtdb
• Skani version: 0.3.1 (pinned, same binary as cdm_skani)
• Entrypoint: skani (no subcommand) - append search (typical) or dist as the first argument

Reference data

Bound at CTS registration time to the existing GTDB-Tk R232 bundle:

• cts-refdata/gtdbtk/r232/gtdbtk_r232_data.tar.gz
• Refdata UUID: bb6352b4-b86f-4e3d-a858-4bc77327ab13
• Mounted at: /ref_data/

Inside the unpacked R232 bundle the skani sketch directory is at /ref_data/release232/skani/database/. Pass that path to skani search -d.

Why R232 and not skani's pre-sketched R226

skani publishes a pre-sketched GTDB at http://faust.compbio.cs.cmu.edu/skani-files/skani_gtdb_r226-v0.3.tar.gz, but that one is R226 (older). The KBase inner-loop stack standardized on R232 for cdm_gtdbtk, and we chose to reuse the same R232 bundle for cdm_skani_gtdb so that any downstream table joining taxonomy (gtdbtk) and nearest-reference ANI (skani_gtdb) on the same genome resolves to the same GTDB version. Closest-reference IDs from skani therefore match closest_genome_reference from gtdbtk by construction.

Version pin (skani 0.3.1)

The skani binary is copied out of ecogenomic/gtdbtk:2.7.2, which is exactly the binary used by GTDB-Tk 2.7.2 to build the R232 skani sketches inside the bundle. Sketch-format compatibility across skani versions is documented as "use the same version that built the database"; pinning here removes that failure mode.

Usage

Typical: search query genomes against the bundled GTDB R232 sketch DB

job = tscli.submit_job(
    "ghcr.io/kbaseincubator/cdm_skani_gtdb:0.1.0",
    input_genomes,                       # one or more user .fna / .fna.gz
    "cts/io/<user>/output/skani_gtdb/run1",
    cluster="kbase",
    declobber=True,
    output_mount_point="/out",
    args=[
        "search",
        "-d", "/ref_data/release232/skani/database/",
        "-o", "/out/hits.tsv",
        "-t", "4",
        "-n", "10",                       # top-10 nearest GTDB references per query
        "--short-header",
        tscli.insert_files(),
    ],
    num_containers=1,
    cpus=4, memory="32GB", runtime="PT1H",
)

Memory: per the skani docs, querying ~140k GTDB representatives takes "seconds with a single processor and ~6 GB of RAM" - the 32 GB budget here is comfort margin.

Output columns

skani search emits a TSV with: Ref_file, Query_file, ANI, Align_fraction_ref, Align_fraction_query, Ref_name, Query_name. With --detailed, also adds contig N50 and other diagnostics. With --ci, adds 5%/95% ANI confidence intervals.

For nearest-neighbor queries the Ref_file basename is the GTDB representative genome ID (e.g. GCA_000147015.1_genomic.fna.gz) - strip _genomic.fna.gz to recover the GTDB ID that matches gtdbtk.bac120.summary.tsv:closest_genome_reference.

Reference

Shaw & Yu, Nature Methods (2023), DOI 10.1038/s41592-023-02018-3.