CheckU

CheckU evaluates bacterial and archaeal genomes with the UNI56 universal single-copy marker set. The program reads amino acid FASTA files or nucleotide assemblies, calls genes with Pyrodigal when needed, and scores markers with PyHMMER. Results include raw completeness, calibrated completeness, contamination, and per-marker hit tables.

Requirements

• FASTA inputs in plain or gzip form (.faa, .fa, .fna, and friends)

Installation (Recommended)

Make sure you have Pixi installed:

curl -fsSL https://pixi.sh/install.sh | sh

Install CheckU with Pixi:

pixi global install \
  -c conda-forge \
  -c bioconda \
  -c https://repo.prefix.dev/astrogenomics \
  checku

Quick test

Small test data sets ship with CheckU. After installation you can confirm the pipeline by running:

checku test

See the Expected Results section below for the expected output tables.

Alternative: pip (PyPI)

pip install checku

Developer install (Pixi)

If you want to download the code and develop locally:

git clone https://github.com/juanvillada/checku
cd checku
pixi install

Quick check

checku --help

If you are running from the repository with Pixi:

pixi run python -m checku --help

You should see the command line help without errors.

Input rules

• Provide either a single FASTA file or a directory of FASTA files.
• Protein files are used as-is. Nucleotide files trigger Pyrodigal gene calls.
• Compressed files (.gz) are supported; they are unpacked into the run workspace.

Running the pipeline

If you are running from the repository with Pixi, replace checku below with pixi run python -m checku.

The examples below use the bundled test data from a source checkout. Replace the paths with your own FASTA inputs, or run checku test after installation.

Pipeline overview

The diagram below shows the main stages executed by CheckU.

graph TD
    A([Start run]) --> B[Collect FASTA inputs from file or directory]
    B --> C[Materialize gzipped files under `work/` when needed]
    C --> D{Detect sequence type}
    D -->|Protein| E[Use supplied protein FASTA]
    D -->|Nucleotide| F[Predict proteins with Pyrodigal]
    F --> E
    E --> G[Search UNI56 HMMs with pyhmmer]
    G --> H[Aggregate marker hits and completeness statistics]
    H --> I[Write `checku_summary.tsv`]
    H --> J[Write `details/checku_presence.tsv`]
    H --> K[Write raw hit tables in `details/hits/`]
    H --> L[Update checkpoint data and logs]
    H -.-> M[Optional: delete predicted proteins when `--clean-intermediate`]
    I --> N([Pipeline complete])
    J --> N
    K --> N
    L --> N
    M --> N

Loading

Single proteome

checku run \
  checku/data/test_genomes/faa/IMGI2140918011.faa \
  --output-dir tmp/proteome_example \
  --cpus 4

Directory of proteomes

checku run \
  checku/data/test_genomes/faa \
  --output-dir tmp/proteome_batch \
  --cpus 8

For directories containing many bins, you can parallelize across genomes:

checku run \
  bins_fna/ \
  --output-dir tmp/high_throughput_batch \
  --cpus 16 \
  --genome-workers 16 \
  --clean-intermediate

--cpus is the total CPU budget for the run. When --genome-workers is greater than 1, CheckU divides that budget across concurrent genome workers and pyhmmer threads per worker. For large directories of MAGs, this is usually much faster than processing one genome at a time.

Single assembly

checku run \
  checku/data/test_genomes/fna/IMG2140918011.fna \
  --output-dir tmp/assembly_example \
  --cpus 4 \
  --clean-intermediate

Use --clean-intermediate if you do not need the predicted protein FASTA after the run.

CheckU-Cal output

Every checku_summary.tsv now contains both:

• completeness — the raw UNI56 marker fraction (markers_detected / 56 * 100)
• completeness_calibrated — a bundled residual-corrected estimate (CheckU-Cal)

The bundled calibration table is used automatically. If you have genome-level metadata available, you can refine the lookup with:

• --calibration-metadata <metadata.tsv> — optional TSV/CSV with genome_id plus either taxonomy_group or GTDB fields such as classification_gtdbtk, gtdbtk_domain, and gtdbtk_phylum
• --calibration-table <table.tsv> — override the bundled manuscript-derived calibration table

When GTDB-style phylum metadata are provided, CheckU-Cal now uses exact phylum-specific residual tables where the shredded benchmark supports them. If an exact phylum/bin combination is too sparse, CheckU falls back to a coarse phylum grouping and then to domain-wide/global residual tables. This keeps the software-facing calibration more specific than the manuscript display groups while retaining safe fallbacks for poorly represented taxa.

Calibration metadata template

The genome_id column must match the CheckU genome identifier, which is derived from the input filename stem. For example, bin_001.fna.gz becomes bin_001.

Minimal example:

genome_id	gtdbtk_domain	gtdbtk_phylum	classification_gtdbtk
bin_001	d__Bacteria	p__Planctomycetota	d__Bacteria;p__Planctomycetota;c__Planctomycetes;o__Planctomycetales;f__Planctomycetaceae;g__Planctomyces;s__
bin_002	d__Archaea	p__Thermoproteota	d__Archaea;p__Thermoproteota;c__Nitrososphaeria;o__Nitrososphaerales;f__Nitrososphaeraceae;g__Nitrososphaera;s__
bin_003	d__Bacteria	p__Bacillota_A	d__Bacteria;p__Bacillota_A;c__Clostridia;o__Eubacteriales;f__Lachnospiraceae;g__Roseburia;s__

If you already maintain a curated calibration stratum, you can also provide taxonomy_group directly:

genome_id	domain	taxonomy_group
bin_001	Bacteria	Planctomycetota
bin_002	Archaea	Thermoproteota
bin_003	Bacteria	Bacillota_A

Accepted metadata columns:

• genome id: genome_id, record_id, taxon_oid, genome, name
• domain: domain, gtdbtk_domain, gtdb_domain
• taxonomy group: taxonomy_group
• phylum: gtdbtk_phylum, gtdbtk__phylum, gtdb_phylum, phylum_gtdb, phylum
• full GTDB classification: classification_gtdbtk, classification, gtdb_classification, gtdb_taxonomy

Resolution order:

1. Use taxonomy_group directly if provided.
2. Otherwise derive the exact phylum from the phylum or GTDB classification fields.
3. If that exact phylum is not represented in the calibration table for the relevant completeness bin, fall back to the coarser group.
4. If taxonomy is absent or unsupported, fall back to domain-wide and then global calibration.

Example run with metadata:

checku run \
  /path/to/mags \
  --calibration-metadata metadata.tsv \
  --output-dir tmp/checku_with_calibration \
  --cpus 8

The summary and presence tables also report the calibration provenance:

• calibration_domain
• calibration_taxonomy_group
• calibration_checku_bin
• calibration_n_train

Custom marker sets

• The default marker file ships with CheckU (UNI56).
• Point --hmm to a different GA-calibrated .hmm file or to a directory that holds .hmm or .hmm.gz profiles.
• Every profile must define GA cutoffs. The run stops early if a profile is missing them or if names are duplicated.

Example:

checku run \
  /path/to/genomes \
  --hmm /path/to/custom_markers.hmm \
  --output-dir tmp/custom_markers \
  --cpus 8

Outputs

All outputs live in the chosen --output-dir.

• checku_summary.tsv — per-genome summary with raw completeness, calibrated completeness, contamination, duplicate counts, calibration provenance, and Pyrodigal gene statistics.
• details/checku_presence.tsv — marker presence/absence matrix.
• details/hits/*.tsv — raw pyhmmer hits with domain scores.
• checkpoint/checku_checkpoint.json — resume data for interrupted runs.
• logs/checku.log — timestamps, command line, and status messages.
• Output tables and logs record input/output locations using absolute paths for reproducibility.

Resume and logging

• Runs resume automatically when --resume is left on (default).
• Use --no-resume to start fresh; the older checkpoint is copied aside.
• Increase --log-level to DEBUG when you need extra detail.

Verification step

Small test data sets ship with CheckU. After installation you can confirm the pipeline by running:

checku test

The command should finish without errors and produce the summary and presence tables described above.

If you are running from the repository with Pixi:

pixi run python -m checku test

Expected results (Bundled test data)

The tables below summarize the expected checku_summary.tsv values for the bundled FAA and FNA test sets. Absolute paths (input/protein columns in the real table) are omitted for privacy.

The real output table also includes calibration_domain, calibration_taxonomy_group, calibration_checku_bin, and calibration_n_train.

FAA (protein inputs):

genome_id	markers_detected	completeness	completeness_calibrated	duplicated_markers	contamination
IMGI2140918011	55	98.21	89.20	0	0.00
IMGI2645727657	56	100.00	89.09	0	0.00
IMGI651324087	56	100.00	89.09	0	0.00
IMGM3300027739_BIN74	36	64.29	57.07	0	0.00
SCISO2808607008	55	98.21	89.20	1	1.79
SDISOGCA_003484685.1	47	83.93	70.34	1	1.79
SHISO2654587767	55	98.21	89.20	1	1.79
SLISOGCF_900639865.1	56	100.00	89.09	1	1.79
SRISO640427127	52	92.86	81.67	0	0.00
SXGCA_000019745.1	55	98.21	89.20	0	0.00
SXGCA_902860225.1_Azoamicus_ciliaticola	51	91.07	79.88	0	0.00
SXISO642555114	54	96.43	87.42	1	1.79

FNA (nucleotide inputs with Pyrodigal):

genome_id	markers_detected	completeness	completeness_calibrated	duplicated_markers	contamination	pyrodigal_genes	pyrodigal_contigs
IMG2140918011	56	100.00	89.09	0	0.00	2974	78
IMG2645727657	56	100.00	89.09	0	0.00	1516	1
IMG2645727657_HALF	46	82.14	68.55	0	0.00	821	1
IMG651324087	56	100.00	89.09	0	0.00	2572	73