codonyat

codon-yat — A codon-aware amino acid variant typer from SAM alignments of viral NGS data.

A Python package for codon-level amino acid variant calling from SAM alignments of viral amplicon sequencing, including:

• SAM parsing that honors amplicon labels and strand orientation.
• CIGAR-aware codon extraction that correctly handles insertions, deletions, and soft clips.
• Ratio balancing, entropy tracking, and TSV/XML exporters mirroring the legacy Perl outputs.
• Configurable CLI flags for strand ratio bounds, entropy sensitivity, and target protein.
• Opt-in research statistics (Wilson CI, strand-bias test, diversity indices) and AAVF v1.0. Research use only.

Highlights

• Produces consistent TSV/XML diagnostics while exposing Python objects (SamContainer, FullReference, etc.) that downstream tooling can import.
• Validates every input file before parsing to surface malformed SAM/FASTA/amplicon data early.
• Logs per-position entropy and strand balance for easier debugging in CI or local runs.
• Supports any protein target via --protein (one name, a comma-separated list, or all; defaults to RT).

Installation

pip install codonyat

Requires Python 3.10 or newer. The only runtime dependency is Biopython.

Running it in a workflow

This repository contains only the Python package. To run codonyat on many samples (from FASTQ or SAM, with read QC, trimming, alignment and a merged report), use the separate Nextflow pipeline nf-codonyat, which installs this package from PyPI.

CLI usage

Once installed, the codonyat entry point is available:

codonyat sample.sam reference.fasta amplicons.tsv

For multi-record references or complex annotations:

codonyat sample.sam reference.fasta amplicons.tsv --annotation annotations.tsv --protein S

Options

Flag	Default	Description
--annotation	None	Optional TSV annotation file (record, protein, description, intervals, strand)
--protein	RT	Protein(s) to analyse: one name, a comma-separated list (e.g. PR,RT,INT, counted in a single pass) or all (every protein annotated in the reference header or TSV). Names must match the FASTA header annotations
--ratio-upper	3.162	Upper bound for strand-ratio balancing
--ratio-lower	0.316	Lower bound for strand-ratio balancing
--entropy-threshold	0.0	Minimum Shannon entropy to mark a position as diverse
--min-baseq	off	Count a codon only if all 3 bases have Phred-like quality ≥ this value
--stats	off	Optional stats TSV (CI, strand-bias, flags, diversity)
--aavf	off	Optional AAVF v1.0 path (no CAT/SRVL)
--min-freq	0.01	Frequency floor (fraction) for lowfreq / AAVF FILTER
--min-coverage	10	Coverage floor for lowcov / AAVF FILTER
--sb-alpha	0.05	Fisher p-value cutoff for the sb flag

Output

The CLI writes two files alongside the input SAM:

• [sam-file].tsv — TSV with columns: FILE, REFERENCE, [RECORD], PROTEIN, VARIANT, POSITION, FREQ, FWCOV, RVCOV, TOTALCOV, RATIO
  • The RECORD column is added only for multi-record references (v1.2.0+)
  • Single-record outputs remain byte-identical to v1.1.0
  • Rows are ordered by genomic position (codon index) along the protein's intervals
  • For multi-interval proteins, this follows the biological reading order through all segments
• [sam-file].xml — per-position <Depth>, <FwCover>, <RvCover>, <Variants>
  • With several proteins, each protein's positions are wrapped in <Protein name="…">; single-protein XML is unchanged

Default TSV/XML are byte-identical to 1.2.1 when the opt-in flags are omitted.

Research statistics (opt-in)

--stats PATH writes a sidecar TSV with LEVEL = variant, codon, or gene:

• variant: frequency as a fraction, Wilson 95% CI, Fisher strand-bias p, flags (lowfreq, sb, lowcov)
• codon / gene: Shannon entropy (raw and normalised), Gini–Simpson, nucleotide π, amino-acid π

--aavf PATH writes AAVF v1.0. FILTER is taken from the flags above. The file does not contain resistance or surveillance fields. Formulas and citations: docs/methods.md.

codonyat-runner reads.sam reference.fasta amps.tsv \
  --protein all --stats sample.stats.tsv --aavf sample.aavf

Reference FASTA format

The reference FASTA supports both single-record and multi-record files. Protein annotations can be provided either in the FASTA header or via a separate TSV file.

FASTA header annotation (backward compatible)

Protein annotations in the format Name(Description):start-end, separated by semicolons:

>K03455|HIVHXB2CG PR(Protease):2253-2549;RT(Reverse Transcriptase):2550-3869;INT(Integrase):4230-5093
TGGAAGGGCTAATTCACTCCCAACGAAGAC...

Multi-interval proteins

For proteins spanning multiple intervals (frameshift, splicing, or circular origin-crossing), use comma-separated intervals. All coordinates are 1-based inclusive:

>MN908947.3 ORF1ab(Polyprotein):266-13468,13468-21555
ATTAAAGGTTTATACCTTCCCAGGTAACAA...

The interval 266-13468,13468-21555 represents a -1 programmed ribosomal frameshift where base 13468 is read twice. Semicolons separate different proteins in the header.

TSV annotation format (v1.2.0+)

For complex annotations, use a TSV file with the --annotation flag:

record	protein	description	intervals	strand
MN908947.3	S	Spike glycoprotein	21563-25384	+
MN908947.3	ORF1ab	Polyprotein	266-13468;13468-21555	+
NC_003977.2	P	DNA polymerase	2307-3221;1-1623	+

Important: Minus-strand genes are not supported and will be rejected with a clear error message.

Package API

Import aa_caller to reuse the core objects:

from aa_caller import SamContainer, FullReference, parse_amplicons

Python wrapper

Use call_variants to run variant calling from Python without touching the CLI:

from pathlib import Path
from aa_caller import call_variants

result = call_variants(
    sam_path=Path("reads.sam"),
    reference_path=Path("reference.fasta"),
    amplicons_path=Path("amps.tsv"),
    protein="PR,RT,INT",  # or "all", "RT", or ["PR", "RT"]
    ratio_upper=3.2,
)

print(result.csv_path, result.xml_path)
print(result.container.variants.keys())

If you already have an argparse.Namespace or mapping of the CLI arguments, call_variants_from_args adapts them directly:

from argparse import Namespace

args = Namespace(
    sam_file="reads.sam",
    reference_file="reference.fasta",
    amplicons_file="amps.tsv",
    protein="RT",
    ratio_upper=3.2,
)

call_variants_from_args(args)

CLI wrapper

A lightweight runner at codonyat-runner exposes the same arguments as call_variants_from_args:

codonyat-runner reads.sam reference.fasta amps.tsv --protein PR,RT,INT --ratio-upper 3.1 --csv-path results.tsv

Development

Install the repository with the optional dev tooling so your local environment matches CI:

pip install --upgrade pip
pip install -e .[dev]

Now you can run the same checks as .github/workflows/python-tests.yml:

ruff check .
python -m pytest
python -m build && twine check dist/*

Project structure

codonyat/
├── aa_caller/            # Python package
│   ├── cli.py            # CLI entry point (codonyat)
│   ├── runner.py         # Python API (call_variants, call_variants_from_args) and codonyat-runner
│   ├── container.py      # SamContainer — variant aggregation engine
│   ├── sam.py            # SamEntry — CIGAR-aware SAM record parser
│   ├── reference.py      # FullReference — FASTA + protein annotation loader
│   ├── models.py         # Protein, Amplicon, Variant, Qual dataclasses
│   ├── validators.py     # Input file validation
│   ├── genetic_code.py   # Codon translation table
│   ├── constants.py      # Default thresholds
│   ├── metrics.py        # Wilson CI, Fisher, diversity indices
│   ├── stats_writer.py   # --stats TSV
│   └── aavf.py           # AAVF v1.0 writer
├── tests/                # pytest suite (unit + integration)
├── docs/index.html       # Project page (GitHub Pages)
├── docs/methods.md       # Formulas and citations for 1.3.0 outputs
└── pyproject.toml        # Package metadata

Releasing

Publishing to PyPI is done by .github/workflows/publish-pypi.yml when a GitHub release is published:

1. Set version in pyproject.toml (e.g. 1.0.3) and merge it to main.
2. Create a tag v<version> and a GitHub release from it.
3. The workflow builds the sdist and wheel, runs twine check, checks that the tag matches the package version, and publishes via PyPI trusted publishing (GitHub environment pypi).

License

MIT