InterPro and Pfam: Domains and Families

By the end of this lesson you will be able to read a protein not as one long string of amino acids but as an assembly of reusable, named parts — its domains — and use InterPro and Pfam to name those parts for BRCA1, closing the loop from its sequence to its 3D structure to what it actually does.

A protein is built from parts, not written from scratch

Evolution rarely invents a new protein letter by letter. It copies, shuffles, and recombines pieces that already work. The unit it reuses is the domain: a stretch of a protein that folds up on its own and carries out one job — bind DNA, grip a zinc ion, dock onto a partner. The same domain turns up in hundreds of otherwise unrelated proteins, because a part that folds and functions independently is worth reusing.

A protein family, then, is a set of proteins (or domains) that descend from a common ancestor and share that inherited piece. Group the sequences by their shared parts and you get something far more useful than a raw sequence: a functional parts list. That grouping is a derived activity, sitting on top of the sequence archives you met in Primary vs Derived Databases — nobody deposits a "family", it is inferred.

Pfam: families defined by profile HMMs

Pfam is the classic family database. Each Pfam entry is not a single reference sequence but a statistical model of a whole family: a profile HMM (profile Hidden Markov Model) — a position-by-position fingerprint that records, for every column of a curated multiple alignment, which amino acids are likely there and how tolerated insertions or deletions are. Score a new protein against that model and you get a match with an E-value, the same evidence-strength number BLAST reports for a sequence hit: a low E-value means the family fit is real, not chance.

Every Pfam family has a stable accession of the form PFxxxxx. Two matter for our story: PF00097 (the zf-C3HC4 RING-finger zinc finger) and PF00533 (the BRCT domain). One change worth noting: the standalone Pfam website was decommissioned in 2022, and Pfam now lives inside InterPro. The accessions and models are unchanged — you just reach them through a new front door.

InterPro: one integration over many signature databases

Pfam is not the only group modelling families. Others — PROSITE, SMART, CDD, PANTHER, PRINTS, and more — each describe domains their own way, and they overlap and disagree. InterPro is the referee. It integrates these member databases, merges signatures that describe the same biological entity into a single InterPro entry, and gives each a stable IPRxxxxxx accession. So instead of several partial answers you get one consolidated call, with every supporting signature listed underneath.

InterProScan is the companion tool: give it a protein sequence and it runs every member database at once, then reports the InterPro-level result.

BRCA1's architecture: a RING and two BRCTs

Now the through-line. The protein you met as UniProt P38398 in UniProt: Swiss-Prot vs TrEMBL is 1863 residues long, but its function concentrates in two domains at its ends. At the N-terminus sits a RING finger (InterPro IPR001841, Pfam PF00097) — a zinc-binding domain that pairs BRCA1 with its partner BARD1 to form an enzyme that tags other proteins for regulation. At the C-terminus sit two tandem BRCT domains (InterPro IPR001357, Pfam PF00533) — phosphopeptide-reading modules that let BRCA1 dock onto damage signals during DNA repair.

This is exactly why the PDB structures from the last lesson are what they are: 1JM7 solves the RING/BARD1 pair, 1JNX solves the BRCT tandem. Crystallographers do not solve 1863 residues at once — they solve domains. Sequence names the parts (InterPro), structure shows one part folded (PDB), and function follows from the fold. Many clinical BRCA1 variants land inside these two domains precisely because that is where the work happens.

Pulling domains with code

You do not have to click. InterPro serves a REST API that returns the domain hits for a UniProt accession as JSON:

curl "https://www.ebi.ac.uk/interpro/api/entry/all/protein/uniprot/P38398/" \
  -H "Accept: application/json"

Or in Python, to list which InterPro entries fall on BRCA1:

import requests

url = "https://www.ebi.ac.uk/interpro/api/entry/all/protein/uniprot/P38398/"
data = requests.get(url, headers={"Accept": "application/json"}).json()

for hit in data["results"]:
    meta = hit["metadata"]
    print(meta["accession"], meta["type"], meta["name"])
    # e.g. IPR001357 domain BRCT ; IPR001841 domain Znf_RING

Run that and BRCA1 stops being an opaque 1863-letter sequence and becomes what it truly is: a RING joined to a pair of BRCTs, each a reusable part InterPro can name in thousands of other proteins too. That is the payoff of domain databases — they turn a private sequence into a public vocabulary.