Sequence Format Conversion

Format conversion looks trivial — call SeqIO.convert and you're done — but it isn't. FASTA is sequence-only; GenBank/EMBL/INSDC carry annotation (features, qualifiers, references); FASTQ carries quality. Crossing those boundaries is a one-way trip. This skill is the safe path: which conversions are lossless, which lose annotation, and how to batch the safe ones at scale.

When to use

• Switching between FASTA and GenBank for a tool that requires one or the other.
• Converting a directory of .gb → .fasta for BLAST.
• Converting EMBL → GenBank for NCBI submission compatibility.
• Stripping quality from FASTQ for tools that need FASTA.

When NOT to use

• Format conversions that need re-validation (e.g., a GenBank you got from a third party — see the bio-format-validation skill from read-qc).
• Lossy round-trips where you actually need the annotation back: don't go FASTA → GenBank and expect features.

Prerequisites

• biopython>=1.83
• For large-scale batch: seqkit (C) is far faster than Python.

Core workflow

1. Decide if the conversion is lossless. Cross-format only when features don't matter, or when both formats are annotation-aware.
2. Use SeqIO.convert(src, src_fmt, dst, dst_fmt) — it returns the record count and handles buffered I/O.
3. For streaming, prefer parse + write when you need to filter or modify on the way through.
4. Always set explicit formats — never rely on extension sniffing alone.

Lossless vs. lossy

Code patterns

GenBank → FASTA (annotation dropped)

from Bio import SeqIO

n = SeqIO.convert("input.gb", "genbank", "out.fasta", "fasta")
print(f"Converted {n} records (annotation discarded)")

FASTA → GenBank (skeleton record, no features)

from Bio import SeqIO
from Bio.SeqRecord import SeqRecord
from Bio.Seq import Seq

recs = []
for r in SeqIO.parse("input.fasta", "fasta"):
    recs.append(SeqRecord(Seq(str(r.seq)), id=r.id, description=r.description,
                          annotations={"molecule_type": "DNA"}))

SeqIO.write(recs, "out.gb", "genbank")

Note: NCBI's tbl2asn will reject records without molecule_type and other minimal metadata. Add a translation table annotation if you'll submit.

EMBL → GenBank

from Bio import SeqIO

n = SeqIO.convert("input.embl", "embl", "out.gb", "genbank")
print(f"Converted {n} records (qualifiers mapped to GenBank qualifiers)")

EMBL → GenBank is lossy in edge cases (e.g., /translation table differences); always re-validate before submission.

FASTQ → FASTA (drop quality)

from Bio import SeqIO
SeqIO.write(SeqIO.parse("reads.fastq", "fastq"), "reads.fasta", "fasta")

FASTA → FASTQ (dummy Q-scores)

There's no canonical way — FASTQ requires quality. Common convention: write a constant Q40 (ASCII I):

from Bio import SeqIO
from Bio.SeqRecord import SeqRecord
from Bio.Seq import Seq

def fasta_to_fastq_q40(in_fa: str, out_fq: str):
    with open(out_fq, "w") as out:
        for r in SeqIO.parse(in_fa, "fasta"):
            quals = [40] * len(r.seq)
            r.letter_annotations["phred_quality"] = quals
            SeqIO.write(r, out, "fastq")

Batch convert a directory

from pathlib import Path
from Bio import SeqIO

for gb in sorted(Path("genbank").glob("*.gb")):
    out = Path("fasta") / (gb.stem + ".fasta")
    n = SeqIO.convert(str(gb), "genbank", str(out), "fasta")
    print(f"{gb.name} -> {out.name}: {n}")

Streaming with a filter (length ≥ 500 bp)

from Bio import SeqIO

def long_records(path, fmt, min_len=500):
    for r in SeqIO.parse(path, fmt):
        if len(r.seq) >= min_len:
            yield r

SeqIO.write(long_records("input.gb", "genbank"), "long_only.fasta", "fasta")

Re-encode legacy FASTQ to Phred+33

from Bio import SeqIO

records = SeqIO.parse("legacy.fastq", "fastq-illumina")
SeqIO.write(records, "modern.fastq", "fastq")

Common pitfalls

• GenBank round-trip loses qualifiers not mapped 1-to-1. Always diff before assuming a clean round-trip.
• SeqIO.convert returns 0 silently if the source format is wrong. Always assert n > 0 in pipelines.
• FASTQ → FASTA can produce files with non-IUPAC characters if the FASTQ had ambiguity codes. Decide whether to clean first.
• molecule_type annotation missing → GenBank writers will warn or reject.
• Don't write features onto a Seq from a SeqIO.parse(... "fasta") record — the seq is a Seq but the record's features list is empty by design.

Validation

• After conversion, grep -c '^>' out.fasta equals the record count.
• For GenBank → FASTA → GenBank, compare Bio.SeqIO.parse() record counts and IDs.
• For FASTQ re-encoding, re-parse with fastq and confirm letter_annotations["phred_quality"] are all 40 in the dummy case.

Open alternatives

References

• Biopython convert formats table: https://biopython.org/wiki/SeqIO#File_Formats
• INSDC feature table: https://www.insdc.org/submitting-standards/feature-table/
• Companion: ors-bioinformatics-sequence-batch-sequence-processing, ors-bioinformatics-sequence-read-write-sequences.

Changelog

• 1.0.0 (2026-06-10): Initial adaptation by Pradyumna Jayaram from bio-format-conversion (bioSkills-main/sequence-io/format-conversion).