pefftacular

A pure-Python library for reading and writing PEFF (PSI Extended FASTA Format) files — the proteomics community's FASTA extension for carrying rich per-entry annotations (PTMs, sequence variants, processed forms, and more) directly in the sequence header. It's for anyone building or consuming protein sequence databases that need more than a bare FASTA header can hold.

pefftacular parses PEFF into typed, structured objects instead of leaving you to regex the header yourself, and it's permissive by default: malformed-but-recoverable files still parse, with spec violations reported as warnings rather than hard failures.

Documentation: https://tacular-omics.github.io/pefftacular/

Highlights

• Zero dependencies — pure Python, nothing else to install.
• Two ways to read — read_peff for the whole file at once, PeffReader to stream entries lazily without loading everything into memory.
• Rich annotations as typed data — variants, UniMod/PSI-MOD modification sites, processed forms, and header-declared custom keys all come back as structured fields, not strings you parse yourself.
• Permissive reading, strict when you want it — spec violations are reported through PeffWarning (opt in to warnings.simplefilter("error", PeffWarning) for strict parsing) while PeffParseError still carries a line, context, and a repair hint for structural failures.
• Round-trip safe writing, including the escaping rules PEFF requires in description lines.
• Shares its API shape with fastatacular, the plain-FASTA sibling library, so switching formats doesn't mean relearning the interface.

Install

pip install pefftacular

Dev install:

just install

Quick start

read_peff — load everything into memory at once:

from pefftacular import read_peff

header, entries = read_peff("proteins.peff")

for entry in entries:
    print(entry.db_unique_id, entry.pname, len(entry.sequence))

PeffReader — iterate lazily without loading the full file:

from pefftacular import PeffReader

with PeffReader("proteins.peff") as reader:
    file_header = reader.header
    for entry in reader:
        process(entry)

Both accept gzip, bzip2 and xz compressed files (proteins.peff.gz) directly, detected from the magic bytes. read_peff, PeffReader, to_records and write_peff take a pandas-style compression argument: "infer" (the default: magic bytes when reading, the .gz/.bz2/.xz suffix when writing), "gzip", "bz2" or "xz" to force a format (an open handle must then be binary), or None for plain text.

Data model

read_peff and PeffReader yield SequenceEntry objects with these fields:

Field	Type	Description
prefix	str	Database prefix (e.g. sp, tr)
db_unique_id	str	Accession (e.g. P12345)
sequence	str	Amino acid sequence
pname	str | None	Protein name (\\PName=)
gname	str | None	Gene name (\\GName=)
ncbi_tax_id	int | None	NCBI taxonomy ID (\\NcbiTaxId=)
length	int | None	Sequence length (\\Length=)
sv	int | None	Sequence version (\\SV=)
ev	int | None	Entry version (\\EV=)
pe	int | None	Protein existence level (\\PE=)
variant_simple	tuple[VariantSimple, ...]	Simple sequence variants
variant_complex	tuple[VariantComplex, ...]	Multi-residue variants (start, end, new sequence, optional tag)
mod_res_unimod	tuple[ModResUnimod, ...]	UniMod modification sites
mod_res_psi	tuple[ModResPsi, ...]	PSI-MOD modification sites
mod_res	tuple[ModRes, ...]	Other named modification sites
processed	tuple[Processed, ...]	Processed sequence forms
custom_values

Annotations

Variants:

from pefftacular import read_peff

_, entries = read_peff("proteins.peff")
entry = entries[0]

for v in entry.variant_simple:
    print(v.position, v.new_amino_acid, v.tag)
    # e.g. 42, "K", "rs12345"

Modifications (UniMod):

for mod in entry.mod_res_unimod:
    print(mod.positions, mod.accession, mod.name)
    # e.g. (17,), "21", "Phospho"

Modifications (PSI-MOD):

for mod in entry.mod_res_psi:
    print(mod.positions, mod.accession, mod.name)
    # e.g. (17,), "MOD:00696", "phosphorylated residue"

Processed forms:

for proc in entry.processed:
    print(proc.start_pos, proc.end_pos, proc.accession, proc.name)
    # e.g. 1, 24, "PRO_0000012345", "Signal peptide"

Custom keys (declared via # CustomKeyDef= in the header):

When the database header declares a custom key, entry values for that key are parsed using its RegExp / FieldNames / FieldTypes and exposed as typed fields on entry.custom_values. The original item text is preserved in raw for lossless round-trips.

Header excerpt:

# CustomKeyDef=(KeyName=SecondaryStructure|Description="..."|ConceptCURIE=BAO:0000014|RegExp="([0-9]+)\|([0-9]+)\|([A-Za-z]+:[0-9]+)?\|(.+)"|FieldNames=StartPosition,EndPosition,CURIE,Description|FieldTypes=integer,integer,string,string)

Entry usage:

>cu:P00001 \SecondaryStructure=(10|20|ncithesaurus:C47937|Helix)

Access:

ss = entry.custom_values["SecondaryStructure"]
ss[0].fields["StartPosition"]    # 10 (int)
ss[0].fields["Description"]      # "Helix"

Supported FieldTypes are XSD basic types (string, integer, decimal, boolean, date, time) plus enumeration(a|b|c). Coercion failures and enumeration mismatches emit PeffWarning (a UserWarning subclass) and fall back to the raw string. If no RegExp is declared, the value is split on | and zipped with FieldNames.

Other non-standard keys (no CustomKeyDef registered) still land in entry.extra as raw strings:

value = entry.extra.get("MyCustomKey")

FASTA and ProForma

from pefftacular import SequenceEntry

entry = SequenceEntry.from_fasta(
    "sp|P31946|1433B_HUMAN 14-3-3 protein beta/alpha OS=Homo sapiens OX=9606 GN=YWHAB PE=1 SV=3",
    "MTMDKSELVQKAKLAEQAERYDDMAAAMK",
)
entry.prefix, entry.db_unique_id, entry.id   # ("sp", "P31946", "1433B_HUMAN")
header, sequence = entry.to_fasta()          # back to a plain FASTA record

entry.to_proforma()                  # \ModResPsi sites, e.g. "MS[MOD:00046]TK..."
entry.to_proforma(mods="unimod")     # from \ModResUnimod sites
entry.to_proforma(variants=entry.variant_simple[:1])   # with a substitution applied
entry.to_proforma(errors="skip")     # None instead of PeffError (e.g. sites past the end)

These use plain strings (pefftacular has no dependencies). With fastatacular, pass (record.raw_header, record.sequence). See From FASTA and ProForma strings.

Tables with pandas or polars

to_records(source) returns one plain dict per entry, so any data-frame library can take the result directly. pefftacular does not ship or require pandas or polars; install whichever you use. (The test suite runs these examples only when the library is installed.)

import pandas as pd
import polars as pl

from pefftacular import to_records

records = to_records("proteins.peff")   # a path, an open text handle, or entries
df = pd.DataFrame(records)
with_variants = df[df["variant_simple"].notna()]    # entries that have \VariantSimple
human = pl.DataFrame(records).filter(pl.col("ncbi_tax_id") == 9606)

PeffReader.to_records() and SequenceEntry.to_record() give the same dicts. The file header is not included. Every record has these keys, in this order (pefftacular.RECORD_KEYS); an absent value is None:

key	type	from
prefix, db_unique_id	str	the identifier prefix:db_unique_id
id, db_unique_id_key	str	\ID, \DbUniqueId
pname, gname, tax_name, comment	str	\PName, \GName, \TaxName, \Comment
ncbi_tax_id, length, sv, ev, pe	int	\NcbiTaxId, \Length, \SV, \EV, \PE
decoy	bool	\Decoy
variant_simple, variant_complex, mod_res_unimod, mod_res_psi, mod_res, processed, disulfide_bond, proteoform	str	the annotation's PEFF value text, e.g. (12|L)(30|*)
custom_values, extra	str	header-declared custom keys / unknown keys as \Key=value text
sequence	str	the residues

The same fields in fastatacular records have other names where each package follows its own model. Rename these to put both in one frame:

field	fastatacular	pefftacular
database prefix	prefix	prefix
accession	accession	db_unique_id
entry name	entry_name	id
protein name, gene	pname, gname	pname, gname
organism name	os_name	tax_name
taxon id, PE, SV	ncbi_tax_id, pe, sv	ncbi_tax_id, pe, sv
other keys	extra, KEY=value pairs	extra, \Key=value pairs
length, residues	length, sequence	length, sequence

Writing

Build a header and entries, then write:

from pefftacular import DatabaseHeader, FileHeader, SequenceEntry, write_peff

db_header = DatabaseHeader(
    prefix="sp",
    db_name="SwissProt",
    db_version="2024_01",
    number_of_entries=1,
)

file_header = FileHeader(
    peff_version="1.0",
    databases=(db_header,),
)

entry = SequenceEntry(
    prefix="sp",
    db_unique_id="P12345",
    sequence="MKTIIALSYIFCLVFA",
    pname="Example protein",
    gname="EXMP",
)

write_peff(file_header, [entry], "output.peff")

dest can be a file path string, a pathlib.Path, or a text-mode file object. entries can be any iterable, including a generator: it is consumed once, as a stream.

Every written line is parsed back and compared with the entry, so a value that would read back differently raises PeffWriteError instead of corrupting the file. For entries that came unchanged from read_peff, pass verify=False to skip that check (about 3x faster writes):

header, entries = read_peff("proteins.peff.gz")
write_peff(header, (e for e in entries if e.variant_simple), "variants.peff", verify=False)

Error handling

Every exception derives from PeffError (a ValueError subclass), so you can catch any failure with one clause. Parse errors carry structured, actionable detail — .line, .context, and a .hint — and attach the offending text and the hint as exception notes, so they also show up in tracebacks:

from pefftacular import PeffError, PeffParseError, read_peff

try:
    header, entries = read_peff("malformed.peff")
except PeffParseError as e:
    print(e.line)     # 1-based line number where it failed
    print(e.context)  # the exact offending text
    print(e.hint)     # a short suggestion for how to fix it
except PeffError:
    ...               # any other pefftacular failure

write_peff validates its input before writing anything and raises PeffWriteError (also a PeffError), with a .hint and the bad entry's 0-based .index (message prefix Entry N: ), for a missing header or an entry with an empty prefix, db_unique_id or sequence. File-system failures (missing directory, no permission) are not wrapped: they raise the usual OSError (FileNotFoundError, PermissionError).

from pefftacular import PeffWriteError

try:
    write_peff(file_header, entries, "output.peff")
except PeffWriteError as e:  # invalid header/entries
    print(e, e.hint)
except OSError as e:          # could not open/write the file
    print(e)

Spec-violation warnings

Reading is permissive: the data is always returned, but anything that violates a PEFF MUST rule (out-of-range positions, missing required fields, NumberOfEntries mismatches, un-coercible custom values, …) is reported through the PeffWarning category. Promote them to errors when you want strict parsing:

import warnings
from pefftacular import PeffWarning, read_peff

warnings.simplefilter("error", PeffWarning)
header, entries = read_peff("suspect.peff")  # now raises on any spec violation

Logging

The library follows the standard logging convention (it attaches a NullHandler and never configures logging itself). Enable a behavioral trace — useful when scripting or debugging with an AI coding agent:

import logging
logging.basicConfig(level=logging.DEBUG)
logging.getLogger("pefftacular").setLevel(logging.DEBUG)

Milestones (entries read/written) log at INFO; file open, header parse, and entry counts log at DEBUG, under the pefftacular.parser / pefftacular.writer loggers.

Development

Contributor and AI-agent guidance lives in CLAUDE.md. The one command to run before committing is just check (formatting, lint, types, and tests — the same gate CI enforces); just fix auto-applies formatting.

just install      # install dependencies
just check        # format-check + lint + type-check + test (pre-commit gate)
just fix          # auto-fix lint + formatting
just test         # run tests
just test-file tests/test_errors.py   # run a single test file
just cov          # run tests with coverage
just build        # build the package
just clean        # remove cache files

Run just with no arguments to list every recipe.

Citation

If you use pefftacular in research, please cite the archived software release. Machine-readable citation metadata is available in CITATION.cff; GitHub's Cite this repository menu can render it as APA or BibTeX. DOI: 10.5281/zenodo.22925639.

License

MIT

Funding

Supported by NIH grants R01AG077046, R01MH132570, R01MH100175, R01HL165168 and U01AG088679.