Matching peptides¶

Aho--Corasick semantics¶

Prozor builds one automaton from the distinct peptide patterns and searches each protein sequence in a single pass. It returns every occurrence, including:

• a peptide repeated at multiple positions in one protein;
• a short peptide nested inside a longer peptide; and
• overlapping matches.

Coordinates use Python's half-open convention: start is inclusive and end is exclusive. The matched length is therefore end - start.

Duplicate input peptide patterns are removed while preserving their first-seen order. Occurrences are not deduplicated; repeated sites remain separate PeptideAnnotation records.

Backend selection¶

Three backend requests are accepted:

All public matching operations default to auto. Both implementations are installed by Prozor, so a normal installation selects Rust. The pure-Python implementation remains an explicit choice and the fallback when Rust cannot be imported.

from dataclasses import dataclass

from prozor.api import annotate_peptides_streaming


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


result = annotate_peptides_streaming(
    ["PEPTIDE"],
    [Protein("P1", "MPEPTIDEX")],
    backend="auto",
)

print(result.requested_backend)  # "auto"
print(result.resolved_backend)   # "ahocorapy" or "ahocorasick_rs"

Both implementations are required to return the same nested and overlapping matches. The Rust adapter explicitly enables overlapping search.

Case handling¶

Annotation functions are case-sensitive and do not normalize their inputs. Consumers should supply peptide and protein strings in a consistent alphabet. For lower-level use, create_automaton(..., case_sensitive=False) provides case-insensitive matching while returning the original keyword spelling.

Streaming contract¶

annotate_peptides_streaming accepts an iterable of records exposing string id and sequence attributes. The iterable may be consumed only once, which allows direct integration with FASTA readers and database cursors without making Prozor depend on their concrete record types. The resulting occurrence annotations are retained in memory.

Prozor deliberately does not own:

• FASTA parsing or header/accession interpretation;
• ProForma parsing or modified-to-unmodified sequence conversion;
• decoy and contaminant classification;
• enzyme settings or identification filtering; or
• persistence into analysis containers.

Optional N-terminal tryptic context¶

annotate_peptides(..., filter_tryptic=True) keeps matches whose start is the protein N-terminus, follows an initiator methionine, or follows one of the configured prefix residues (by default R or K). This is an N-terminal context filter, not a complete in-silico digestion model.