Prozor

Typed peptide-to-protein matching and deterministic greedy-parsimony protein inference.

This repository is a Python port of the R prozor package.

Prozor offers backend-neutral Aho--Corasick matching over mappings or streaming protein records, occurrence-level annotations, unique peptide--protein edges, and deterministic protein inference. The core deliberately does not parse FASTA files or depend on pandas, AnnData, MuData, MuLink, workflow engines, or consumer CLIs.

Documentation · Getting started · API reference

Documentation: https://anndata-omics-bridge.github.io/prozor/

Installation

python -m pip install prozor

Both matching implementations are installed. The public matching operations default to backend="auto", which selects ahocorasick_rs. The portable ahocorapy implementation remains directly selectable and is the automatic runtime fallback if Rust cannot be imported. Results record both the requested and concrete backend.

Quick start

from dataclasses import dataclass

from prozor.api import annotate_peptides_streaming, greedy_parsimony


@dataclass(frozen=True, slots=True)
class Protein:
    id: str
    sequence: str


matches = annotate_peptides_streaming(
    ["PEPTIDE", "SEQUENCE"],
    [
        Protein("P1", "MYPEPTIDESEQUENCE"),
        Protein("P2", "XXSEQUENCEXX"),
    ],
)

for match in matches:
    print(match.peptide, match.protein_id, match.start, match.end)

edges = {(match.peptide, match.protein_id) for match in matches}
protein_groups = greedy_parsimony(edges)
print(protein_groups.to_dict())

Matches include nested, overlapping, and repeated sites using half-open coordinates. Prozor matches the exact strings supplied by the consumer; FASTA header interpretation, normalization, decoy classification, and persistence remain application policy.

Development

uv sync --group dev --group docs
.venv/bin/pre-commit install --hook-type pre-commit --hook-type pre-push
make check

make check runs Ruff, strict Pyright, dependency validation, branch-coverage tests, a strict documentation build, and wheel/sdist validation. All Python commands run from the synchronized project .venv.

Provenance and license

This implementation ports the algorithmic behavior of the R prozor package and consolidates the matching and inference core for Python consumers such as APB and diann_runner.

The Python distribution is an independent implementation of that algorithm rather than a translation of the R source, and it is licensed under MIT. The R reference package remains GPL-3, and its license does not extend here: the copyright holder of both packages is the same author, and algorithms are not themselves subject to copyright.

MIT keeps the whole anndata-omics-bridge spine under one permissive license, so that consumers such as APB can be redistributed without inheriting copyleft obligations from this package.