Skip to content

Biopython interop

pymafft duck-types Bio.SeqRecord.SeqRecord — anything with .id and .seq attributes works as input. Biopython itself is not a runtime dependency. The cooperation is one-way at the input boundary (pymafft.align reads SeqRecords) and two-way at the output boundary (AlignmentResult produces FASTA that re-parses cleanly).

SeqRecord → pymafft

from Bio import SeqIO
import pymafft

records = list(SeqIO.parse("input.fasta", "fasta"))
result  = pymafft.align(records, strategy="linsi", maxiterate=1000)

The id mapping: SeqRecord.id becomes the alignment row name — NOT .name (which is set to .id if not provided) and NOT .description (which includes the post-id header text). This matches how Biopython itself treats id as the stable identifier.

pymafft → MultipleSeqAlignment

Three routes — pick by style preference, the result is identical.

msa = result.to_biopython()
assert msa.get_alignment_length() == result.width

Each row becomes a SeqRecord with id set to the row name and an empty description. Biopython is imported lazily, inside this call only — pymafft still has no runtime dependency on it, and the method raises ImportError if it is missing.

from io import StringIO
from Bio import AlignIO

msa = AlignIO.read(StringIO(result.to_fasta()), "fasta")
assert msa.get_alignment_length() == result.width

from Bio.Seq import Seq
from Bio.SeqRecord import SeqRecord
from Bio.Align import MultipleSeqAlignment

msa = MultipleSeqAlignment([
    SeqRecord(Seq(s.sequence), id=s.name, description="")
    for s in result.sequences
])

Full round-trip

from Bio import SeqIO, AlignIO
from io import StringIO
import pymafft

# Read → align → write
records = list(SeqIO.parse("opsins.fasta", "fasta"))
result  = pymafft.align(records, strategy="linsi", maxiterate=1000)

AlignIO.write([result.to_biopython()], "opsins.aligned.aln", "clustal")

The alignment options are the CLI's, so the panaroo-style nucleotide call works the same way over SeqRecords:

records = list(SeqIO.parse("cluster.fasta", "fasta"))
result  = pymafft.align(records, strategy="auto", adjust_direction=True,
                        threads=1, seq_type="nuc")

Note that --adjustdirection prefixes reverse-complemented rows with _R_ in the row name, exactly as the command line does.

Header caveat

SeqIO.write(records, ..., "fasta") emits headers of the form id description. If you pass a SeqRecord directly to pymafft.align, the row name is just id. If you serialize the records to FASTA first and then call pymafft.align_fasta_string(...), the row name is the whole header (id description).

When this matters, prefer the direct pymafft.align(records) path — the test suite verifies both produce equivalent aligned sequences, but the names differ.

Tests

The interop guarantees above are exercised by 18 tests in crates/pymafft/tests/test_biopython.py. The suite skips cleanly when Biopython isn't installed (via pytest.importorskip("Bio")).