The NCBI Entrez Ecosystem · Subtopic 2 of 5 · 11 min
RefSeq: Curated Reference Sequences
By the end of this lesson you will know exactly what you are pointing at when you write "the BRCA1 reference is NM_007294" — why that one clean record exists, who vouches for it, and why it belongs in your methods section instead of a raw archive entry you happened to download. GenBank, the nucleotide archive, keeps a permanent record of everything anyone ever submitted, which means the same gene can turn up in dozens of overlapping records — some redundant, some contradictory, none of them "the" answer. RefSeq, NCBI's Reference Sequence collection, is NCBI's fix for that problem.
Derived from INSDC, not part of it
Here is the single most important sentence about RefSeq: it is derived from the archive, not part of it.
Recall from Primary vs Derived Databases the split between a primary database (an archive that stores exactly what was submitted) and a derived database (one that computes a new, curated product on top). GenBank, ENA and DDBJ together form the INSDC — the International Nucleotide Sequence Database Collaboration, the shared archive of raw submissions. RefSeq is the curated layer NCBI builds from those submissions. It reads the archive, then produces one deliberate record per molecule.
So a RefSeq entry does not replace a GenBank entry. It points back at the archive evidence it was built from, and adds a curator's judgment on top.
One clean record per molecule
The design goal of RefSeq is non-redundancy: one reference mRNA, one reference protein, one reference chromosome, per biological molecule per organism. Where GenBank might hold many submitted versions of the BRCA1 transcript, RefSeq gives you exactly one to cite: NM_007294.
That single record is why RefSeq is the backbone of almost every downstream tool. When a variant caller reports a change "in NM_007294", every reader knows precisely which sequence and coordinate system is meant. There is no ambiguity to resolve, because there is only one reference.
You met the RefSeq prefix grammar back in Reading a Prefix; here it is, anchored to our thread gene:
| Prefix | Molecule | BRCA1 example |
|---|---|---|
NM_ | curated mRNA (transcript) | NM_007294 |
NP_ | curated protein | NP_009225 |
NC_ | curated complete genomic | NC_000017 (chromosome 17) |
The underscore is the visible seam: a raw GenBank submission like U14680 never has one, a RefSeq record always does. NM_007294 encodes the protein NP_009225, and both sit on chromosome NC_000017 — molecule type alone tells you their relationship before you read any annotation.
Review status: how much a human touched it
Not every RefSeq record is equally vetted, and RefSeq is honest about it. Each carries a review status — a label saying how far it got through curation.
| Status | What it means |
|---|---|
REVIEWED | A curator examined it against the literature and evidence; the highest tier. |
VALIDATED | Passed automated validation and initial review, not yet fully curated. |
PROVISIONAL | Automatically derived from a submission, awaiting review. |
MODEL | Predicted computationally from a genome annotation (the XM_/XP_ records). |
For a heavily studied gene like BRCA1, NM_007294 sits at REVIEWED. That status is not decoration — it is provenance you can act on. For an obscure gene in a freshly sequenced organism, a MODEL prediction may be all that exists: usable, but flagged in your own mind as provisional.
You can read the status straight from the record with Biopython:
from Bio import Entrez, SeqIO
Entrez.email = "[email protected]" # NCBI requires an identity
with Entrez.efetch(db="nucleotide", id="NM_007294",
rettype="gb", retmode="text") as handle:
record = SeqIO.read(handle, "genbank")
print(record.id) # NM_007294.4 (accession.version)
print(record.annotations["comment"][:40]) # "REVIEWED REFSEQ: ..."Try it yourself: fetch NM_007294 and read the top of the record. The COMMENT block names the review status and, crucially, lists the GenBank accessions the reference was built from — the archive evidence behind the curated record. That is RefSeq telling you its own sources. Note also the .4 after the accession: RefSeq records are versioned, and the number ticks up when curators revise the sequence. Cite the version when reproducibility matters — see Versioning.
Why you cite RefSeq, not raw GenBank
Put it together and the practical rule falls out. When you name a reference in a paper, a variant report, or a pipeline config, you cite the RefSeq accession — NM_007294 — not a raw GenBank entry.
Three reasons hold this up. First, stability of meaning: the RefSeq record is non-redundant, so NM_007294 names one sequence everyone can agree on, while a random GenBank submission is one lab's snapshot from one day. Second, traceable curation: the review status and the source list in the COMMENT block tell a reader how much scrutiny the sequence has already had, before they lean on it. Third, downstream compatibility: clinical variant databases, annotation tools, and genome browsers default to RefSeq accessions, so citing one keeps your work interoperable with everything built on top of it.
That is the whole case for RefSeq in one sentence: it turns "a sequence for BRCA1" into "the sequence for BRCA1," with a review status you can check and a version number you can pin. Reach for NM_007294, not a loose GenBank submission, whenever precision is the point.