SeqBench

How to Open an .embl File (and Why It Looks Like GenBank)

6 min read · Updated November 14, 2026

You download a record from ENA and get a file called something like U49845.dat, or a colleague in Europe sends you an .embl, and nothing you have opens it. Every tutorial assumes GenBank, and the two look different enough at a glance — different line prefixes, no LOCUS, a sequence block that starts with SQ — that it is reasonable to assume they are unrelated formats.

They are not. EMBL, GenBank and DDBJ are three layouts of ONE record, maintained as a single shared specification by the International Nucleotide Sequence Database Collaboration. The feature table — the part you actually want — is byte-for-byte the same grammar in all three. Here is what really differs, and what to do about it.

One record, three layouts

INSDC is not a fourth database, it is the agreement under which the three exchange records nightly. A submitted sequence ends up in all three, and each serves the same record in its own house style: GenBank at NCBI, ENA (the European Nucleotide Archive) in EMBL format, and DDBJ in Japan.

So the feature keys are the same — CDS, gene, promoter, misc_feature, source — the location grammar is the same — 1..2686, complement(70..80), join(...), the angle brackets for a partial feature — and the qualifiers are the same /gene, /product, /note, /translation, /codon_start. This has been checked on real records rather than assumed: U49845 served from ENA and from NCBI carries identical feature keys and locations, a complement(<3300..>4037) included.

The extension follows the same pattern of harmless variation. .embl and .dat are both EMBL flat files, exactly as .gb, .gbk, .genbank, .gbff and .ape are all GenBank flat files. The extension tells your operating system which application to launch; it says almost nothing about what is inside.

The ID line, where GENBANK puts LOCUS

This is the biggest visible difference and it carries the topology. An EMBL record opens with:

ID U49845; SV 1; linear; genomic DNA; STD; FUN; 5028 BP.

In order: the accession, the sequence version, the topology, the molecule type, the division, the taxonomic division and the length. GenBank's LOCUS line carries the same facts in different positions, with the topology after the molecule type.

There are two generations of this line and both are still served. Records written after about 2006 use the semicolon-separated form above. Older ones read ID X56734 standard; RNA; PLN; 1859 BP. — no SV field and no sequence version at all. A reader that only knows one of the two will refuse a large slice of the public archive.

The other header lines map one to one: DE is DEFINITION, AC is ACCESSION, SV gives the version that GenBank writes as VERSION, and KW is KEYWORDS. References, dates, cross-references and the taxonomy block look different and are not needed to read the sequence or draw the map.

The sequence block counts from the other end

The sequence begins after a line like SQ Sequence 5028 BP; 1254 A; 1351 C; 1254 G; 1169 T; 0 other; and it is the one place where the layout difference is a correctness problem rather than a cosmetic one.

EMBL puts the running position at the END of each line. GenBank puts it at the START. Residues are grouped in tens either way, so the letters survive a copy untouched and only the count moves.

That matters because a reader is written for ONE of the two layouts. A GenBank ORIGIN reader given an EMBL line finds no leading count and refuses the block — which is the good failure, and it is what this site's own reader does (checked on a real line rather than assumed). The bad failure is a reader that is tolerant enough to accept both and takes the trailing number as residues, or drops the first block; either produces a record that looks fine and is not the one you sent. Keeping the count in the position the format puts it is what makes the difference visible at all.

It is also the reason to read the length claim two ways: the BP count on the ID line is a claim by the submitter, and the residues between SQ and // are what is actually there. When they disagree, believe the residues.

What conversion actually has to change

What is NOT lost is the annotation, which is the part that matters. The features, their locations and their qualifiers arrive unchanged; a spliced CDS written as join(...) is still one feature with its parts, and a feature marked partial with < or > is still partial.

  • The ID line, into a LOCUS line — and the topology has to survive, because a circular plasmid read as linear loses every fragment size that spans the origin.
  • SQ plus its trailing-count block, into ORIGIN plus a nine-column leading count. This is the one change that is correctness rather than cosmetics.
  • // stays, and so does everything inside the feature table.
  • Everything else — references, dates, DR cross-references, the OC taxonomy block — is dropped, because no reader on this side consumes it.

A record with almost no annotation is normal, in Europe as in the US

The same surprise that greets people opening a GenBank file greets them here. Older database records often carry one feature, a source spanning the whole molecule, saying only what organism the sequence came from. A plasmid map you have seen in a paper is annotation somebody added on top of the sequence; it is not in the deposited record, and it is frequently not in the database entry either.

That is not a reason to distrust the file. It means the map has to be built: the sequence is the real content, the features are a convenience, and a viewer that shows one band around the whole circle is showing you exactly what is in there.

FAQ

Is an .embl file the same as a GenBank file?

Same record, different layout. EMBL, GenBank and DDBJ are the three INSDC member formats, and the feature table — keys, locations and qualifiers — is identical across them. What differs is the header (ID with SV versus LOCUS) and the sequence block, where EMBL puts the running position at the end of each line and GenBank at the start.

Why does my .dat file contain DNA?

Because EMBL uses .dat as one of its flat-file extensions, and it collides with dozens of unrelated formats that use the same letters. If the first line starts with ID followed by an accession, a semicolon and a length in BP, it is an EMBL record. If it does not, look at the first few bytes rather than the extension.

Can I convert EMBL to GenBank without losing the features?

Yes — the feature table needs no translation at all. A conversion has to rewrite the ID line as LOCUS (keeping the topology), re-emit the SQ block as ORIGIN with the count at the front, and add the // terminator. Everything between FEATURES and the sequence block comes across unchanged, including spliced join() locations and partial-feature markers.

Does EMBL store the sequence in a different order?

No — the residues are the same, in the same 5'→3' orientation, grouped in tens. Only the position number's placement differs, at the end of the line instead of the start. That placement is what a conversion has to respect: a reader written for GenBank's layout refuses an EMBL line (no leading count), and a reader lenient enough to accept both can take the trailing number as residues and return a plausible record with the wrong bases.

Related references

Related tools

Related guides

Helpful?